blob: 8cb75a6efec35337830f08c74ccb1a9f4f1d8e1c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 */
33
Joe Perchesa4aee5c2009-12-13 20:06:07 -080034#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/kernel.h>
37#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/types.h>
39#include <linux/fcntl.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040044#include <net/ip.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/ip.h>
Jay Vosburgh169a3e62005-06-26 17:54:11 -040046#include <linux/tcp.h>
47#include <linux/udp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
Stephen Hemminger3d632c32009-06-12 19:02:48 +000056#include <linux/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <asm/system.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include <asm/dma.h>
Stephen Hemminger3d632c32009-06-12 19:02:48 +000059#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/errno.h>
61#include <linux/netdevice.h>
62#include <linux/inetdevice.h>
Jay Vosburgha816c7c2007-02-28 17:03:37 -080063#include <linux/igmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#include <linux/etherdevice.h>
65#include <linux/skbuff.h>
66#include <net/sock.h>
67#include <linux/rtnetlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <linux/smp.h>
69#include <linux/if_ether.h>
70#include <net/arp.h>
71#include <linux/mii.h>
72#include <linux/ethtool.h>
73#include <linux/if_vlan.h>
74#include <linux/if_bonding.h>
David Sterbab63bb732007-12-06 23:40:33 -080075#include <linux/jiffies.h>
Neil Hormane843fa52010-10-13 16:01:50 +000076#include <linux/preempt.h>
Jay Vosburghc3ade5c2005-06-26 17:52:20 -040077#include <net/route.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020078#include <net/net_namespace.h>
Eric W. Biedermanec87fd32009-10-29 14:18:26 +000079#include <net/netns/generic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#include "bonding.h"
81#include "bond_3ad.h"
82#include "bond_alb.h"
83
84/*---------------------------- Module parameters ----------------------------*/
85
86/* monitor all links that often (in milliseconds). <=0 disables monitoring */
87#define BOND_LINK_MON_INTERV 0
88#define BOND_LINK_ARP_INTERV 0
89
90static int max_bonds = BOND_DEFAULT_MAX_BONDS;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +000091static int tx_queues = BOND_DEFAULT_TX_QUEUES;
Ben Hutchingsad246c92011-04-26 15:25:52 +000092static int num_peer_notif = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093static int miimon = BOND_LINK_MON_INTERV;
Stephen Hemminger3d632c32009-06-12 19:02:48 +000094static int updelay;
95static int downdelay;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096static int use_carrier = 1;
Stephen Hemminger3d632c32009-06-12 19:02:48 +000097static char *mode;
98static char *primary;
Jiri Pirkoa5499522009-09-25 03:28:09 +000099static char *primary_reselect;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000100static char *lacp_rate;
stephen hemminger655f8912011-06-22 09:54:39 +0000101static int min_links;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000102static char *ad_select;
103static char *xmit_hash_policy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104static int arp_interval = BOND_LINK_ARP_INTERV;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000105static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106static char *arp_validate;
107static char *fail_over_mac;
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000108static int all_slaves_active = 0;
Stephen Hemmingerd2991f72009-06-12 19:02:44 +0000109static struct bond_params bonding_defaults;
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000110static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
112module_param(max_bonds, int, 0);
113MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
Andy Gospodarekbb1d9122010-06-02 08:40:18 +0000114module_param(tx_queues, int, 0);
115MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
Ben Hutchingsad246c92011-04-26 15:25:52 +0000116module_param_named(num_grat_arp, num_peer_notif, int, 0644);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000117MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
118 "failover event (alias of num_unsol_na)");
Ben Hutchingsad246c92011-04-26 15:25:52 +0000119module_param_named(num_unsol_na, num_peer_notif, int, 0644);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000120MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
121 "failover event (alias of num_grat_arp)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122module_param(miimon, int, 0);
123MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124module_param(updelay, int, 0);
125MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126module_param(downdelay, int, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800127MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
128 "in milliseconds");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129module_param(use_carrier, int, 0);
Mitch Williams2ac47662005-11-09 10:35:03 -0800130MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132module_param(mode, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000133MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
Mitch Williams2ac47662005-11-09 10:35:03 -0800134 "1 for active-backup, 2 for balance-xor, "
135 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 "6 for balance-alb");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137module_param(primary, charp, 0);
138MODULE_PARM_DESC(primary, "Primary network device to use");
Jiri Pirkoa5499522009-09-25 03:28:09 +0000139module_param(primary_reselect, charp, 0);
140MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
141 "once it comes up; "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
144 "better, "
145 "2 for only on active slave "
146 "failure");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147module_param(lacp_rate, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000148MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
149 "0 for slow, 1 for fast");
Jay Vosburghfd989c82008-11-04 17:51:16 -0800150module_param(ad_select, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000151MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
152 "0 for stable (default), 1 for bandwidth, "
153 "2 for count");
stephen hemminger655f8912011-06-22 09:54:39 +0000154module_param(min_links, int, 0);
155MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
156
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400157module_param(xmit_hash_policy, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000158MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
159 "0 for layer 2 (default), 1 for layer 3+4, "
160 "2 for layer 2+3");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161module_param(arp_interval, int, 0);
162MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
163module_param_array(arp_ip_target, charp, NULL, 0);
164MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700165module_param(arp_validate, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000166MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
167 "0 for none (default), 1 for active, "
168 "2 for backup, 3 for all");
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700169module_param(fail_over_mac, charp, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000170MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
171 "the same MAC; 0 for none (default), "
172 "1 for active, 2 for follow");
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000173module_param(all_slaves_active, int, 0);
174MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
Andy Gospodarek90e62472011-05-25 04:41:59 +0000175 "by setting active flag for all slaves; "
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000176 "0 for never (default), 1 for always.");
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000177module_param(resend_igmp, int, 0);
Andy Gospodarek90e62472011-05-25 04:41:59 +0000178MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
179 "link failure");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
181/*----------------------------- Global variables ----------------------------*/
182
Neil Hormane843fa52010-10-13 16:01:50 +0000183#ifdef CONFIG_NET_POLL_CONTROLLER
Neil Hormanfb4fa762010-12-06 09:05:50 +0000184atomic_t netpoll_block_tx = ATOMIC_INIT(0);
Neil Hormane843fa52010-10-13 16:01:50 +0000185#endif
186
Eric Dumazetf99189b2009-11-17 10:42:49 +0000187int bond_net_id __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000189static __be32 arp_target[BOND_MAX_ARP_TARGETS];
190static int arp_ip_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191static int bond_mode = BOND_MODE_ROUNDROBIN;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000192static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
193static int lacp_fast;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800195const struct bond_parm_tbl bond_lacp_tbl[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{ "slow", AD_LACP_SLOW},
197{ "fast", AD_LACP_FAST},
198{ NULL, -1},
199};
200
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800201const struct bond_parm_tbl bond_mode_tbl[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{ "balance-rr", BOND_MODE_ROUNDROBIN},
203{ "active-backup", BOND_MODE_ACTIVEBACKUP},
204{ "balance-xor", BOND_MODE_XOR},
205{ "broadcast", BOND_MODE_BROADCAST},
206{ "802.3ad", BOND_MODE_8023AD},
207{ "balance-tlb", BOND_MODE_TLB},
208{ "balance-alb", BOND_MODE_ALB},
209{ NULL, -1},
210};
211
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800212const struct bond_parm_tbl xmit_hashtype_tbl[] = {
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400213{ "layer2", BOND_XMIT_POLICY_LAYER2},
214{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
Jay Vosburgh6f6652b2007-12-06 23:40:34 -0800215{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400216{ NULL, -1},
217};
218
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800219const struct bond_parm_tbl arp_validate_tbl[] = {
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700220{ "none", BOND_ARP_VALIDATE_NONE},
221{ "active", BOND_ARP_VALIDATE_ACTIVE},
222{ "backup", BOND_ARP_VALIDATE_BACKUP},
223{ "all", BOND_ARP_VALIDATE_ALL},
224{ NULL, -1},
225};
226
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800227const struct bond_parm_tbl fail_over_mac_tbl[] = {
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700228{ "none", BOND_FOM_NONE},
229{ "active", BOND_FOM_ACTIVE},
230{ "follow", BOND_FOM_FOLLOW},
231{ NULL, -1},
232};
233
Jiri Pirkoa5499522009-09-25 03:28:09 +0000234const struct bond_parm_tbl pri_reselect_tbl[] = {
235{ "always", BOND_PRI_RESELECT_ALWAYS},
236{ "better", BOND_PRI_RESELECT_BETTER},
237{ "failure", BOND_PRI_RESELECT_FAILURE},
238{ NULL, -1},
239};
240
Jay Vosburghfd989c82008-11-04 17:51:16 -0800241struct bond_parm_tbl ad_select_tbl[] = {
242{ "stable", BOND_AD_STABLE},
243{ "bandwidth", BOND_AD_BANDWIDTH},
244{ "count", BOND_AD_COUNT},
245{ NULL, -1},
246};
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248/*-------------------------- Forward declarations ---------------------------*/
249
Stephen Hemminger181470f2009-06-12 19:02:52 +0000250static int bond_init(struct net_device *bond_dev);
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +0000251static void bond_uninit(struct net_device *bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
253/*---------------------------- General routines -----------------------------*/
254
Amerigo Wangbd33acc2011-03-06 21:58:46 +0000255const char *bond_mode_name(int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256{
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800257 static const char *names[] = {
258 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
259 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
260 [BOND_MODE_XOR] = "load balancing (xor)",
261 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000262 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800263 [BOND_MODE_TLB] = "transmit load balancing",
264 [BOND_MODE_ALB] = "adaptive load balancing",
265 };
266
267 if (mode < 0 || mode > BOND_MODE_ALB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 return "unknown";
Holger Eitzenberger77afc922008-12-09 23:08:09 -0800269
270 return names[mode];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
272
273/*---------------------------------- VLAN -----------------------------------*/
274
275/**
276 * bond_add_vlan - add a new vlan id on bond
277 * @bond: bond that got the notification
278 * @vlan_id: the vlan id to add
279 *
280 * Returns -ENOMEM if allocation failed.
281 */
282static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
283{
284 struct vlan_entry *vlan;
285
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -0800286 pr_debug("bond: %s, vlan id %d\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800287 (bond ? bond->dev->name : "None"), vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
Brian Haley305d5522008-11-04 17:51:14 -0800289 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000290 if (!vlan)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
293 INIT_LIST_HEAD(&vlan->vlan_list);
294 vlan->vlan_id = vlan_id;
295
296 write_lock_bh(&bond->lock);
297
298 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
299
300 write_unlock_bh(&bond->lock);
301
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -0800302 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
304 return 0;
305}
306
307/**
308 * bond_del_vlan - delete a vlan id from bond
309 * @bond: bond that got the notification
310 * @vlan_id: the vlan id to delete
311 *
312 * returns -ENODEV if @vlan_id was not found in @bond.
313 */
314static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
315{
Pavel Emelyanov0883bec2008-05-17 21:10:10 -0700316 struct vlan_entry *vlan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 int res = -ENODEV;
318
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -0800319 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Neil Hormane843fa52010-10-13 16:01:50 +0000321 block_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 write_lock_bh(&bond->lock);
323
Pavel Emelyanov0883bec2008-05-17 21:10:10 -0700324 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 if (vlan->vlan_id == vlan_id) {
326 list_del(&vlan->vlan_list);
327
Holger Eitzenberger58402052008-12-09 23:07:13 -0800328 if (bond_is_lb(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 bond_alb_clear_vlan(bond, vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800331 pr_debug("removed VLAN ID %d from bond %s\n",
332 vlan_id, bond->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
334 kfree(vlan);
335
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 res = 0;
337 goto out;
338 }
339 }
340
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800341 pr_debug("couldn't find VLAN ID %d in bond %s\n",
342 vlan_id, bond->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
344out:
345 write_unlock_bh(&bond->lock);
Neil Hormane843fa52010-10-13 16:01:50 +0000346 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 return res;
348}
349
350/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 * bond_next_vlan - safely skip to the next item in the vlans list.
352 * @bond: the bond we're working on
353 * @curr: item we're advancing from
354 *
355 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356 * or @curr->next otherwise (even if it is @curr itself again).
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000357 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 * Caller must hold bond->lock
359 */
360struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
361{
362 struct vlan_entry *next, *last;
363
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000364 if (list_empty(&bond->vlan_list))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
367 if (!curr) {
368 next = list_entry(bond->vlan_list.next,
369 struct vlan_entry, vlan_list);
370 } else {
371 last = list_entry(bond->vlan_list.prev,
372 struct vlan_entry, vlan_list);
373 if (last == curr) {
374 next = list_entry(bond->vlan_list.next,
375 struct vlan_entry, vlan_list);
376 } else {
377 next = list_entry(curr->vlan_list.next,
378 struct vlan_entry, vlan_list);
379 }
380 }
381
382 return next;
383}
384
Neil Horman374eeb52011-06-03 10:35:52 +0000385#define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387/**
388 * bond_dev_queue_xmit - Prepare skb for xmit.
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000389 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 * @bond: bond device that got this skb for tx.
391 * @skb: hw accel VLAN tagged skb to transmit
392 * @slave_dev: slave that is supposed to xmit this skbuff
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000394int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
395 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
Ben Hutchings83874512010-12-13 08:19:28 +0000397 skb->dev = slave_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 skb->priority = 1;
Neil Horman374eeb52011-06-03 10:35:52 +0000399
400 skb->queue_mapping = bond_queue_mapping(skb);
401
Amerigo Wang080e4132011-02-17 23:43:33 +0000402 if (unlikely(netpoll_tx_running(slave_dev)))
Amerigo Wang8a8efa22011-02-17 23:43:32 +0000403 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
Amerigo Wang080e4132011-02-17 23:43:33 +0000404 else
WANG Congf6dc31a2010-05-06 00:48:51 -0700405 dev_queue_xmit(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 return 0;
408}
409
410/*
Jiri Pirkocc0e4072011-07-20 04:54:46 +0000411 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
412 * We don't protect the slave list iteration with a lock because:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 * a. This operation is performed in IOCTL context,
414 * b. The operation is protected by the RTNL semaphore in the 8021q code,
415 * c. Holding a lock with BH disabled while directly calling a base driver
416 * entry point is generally a BAD idea.
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000417 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 * The design of synchronization/protection for this operation in the 8021q
419 * module is good for one or more VLAN devices over a single physical device
420 * and cannot be extended for a teaming solution like bonding, so there is a
421 * potential race condition here where a net device from the vlan group might
422 * be referenced (either by a base driver or the 8021q code) while it is being
423 * removed from the system. However, it turns out we're not making matters
424 * worse, and if it works for regular VLAN usage it will work here too.
425*/
426
427/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
429 * @bond_dev: bonding net device that got called
430 * @vid: vlan id being added
431 */
432static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
433{
Wang Chen454d7c92008-11-12 23:37:49 -0800434 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 struct slave *slave;
436 int i, res;
437
438 bond_for_each_slave(bond, slave, i) {
439 struct net_device *slave_dev = slave->dev;
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
442 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800443 slave_ops->ndo_vlan_rx_add_vid) {
444 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 }
446 }
447
448 res = bond_add_vlan(bond, vid);
449 if (res) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800450 pr_err("%s: Error: Failed to add vlan id %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 bond_dev->name, vid);
452 }
453}
454
455/**
456 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
457 * @bond_dev: bonding net device that got called
458 * @vid: vlan id being removed
459 */
460static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
461{
Wang Chen454d7c92008-11-12 23:37:49 -0800462 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 int i, res;
465
466 bond_for_each_slave(bond, slave, i) {
467 struct net_device *slave_dev = slave->dev;
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800468 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800471 slave_ops->ndo_vlan_rx_kill_vid) {
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800472 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
474 }
475
476 res = bond_del_vlan(bond, vid);
477 if (res) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800478 pr_err("%s: Error: Failed to remove vlan id %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 bond_dev->name, vid);
480 }
481}
482
483static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
484{
485 struct vlan_entry *vlan;
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800486 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800489 !(slave_ops->ndo_vlan_rx_add_vid))
Jay Vosburgh03dc2f42010-07-21 12:14:48 +0000490 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800492 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
493 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494}
495
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000496static void bond_del_vlans_from_slave(struct bonding *bond,
497 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498{
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800499 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 struct vlan_entry *vlan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
502 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800503 !(slave_ops->ndo_vlan_rx_kill_vid))
Jiri Pirkocc0e4072011-07-20 04:54:46 +0000504 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
506 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Jay Vosburghf35188f2010-07-21 12:14:47 +0000507 if (!vlan->vlan_id)
508 continue;
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800509 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511}
512
513/*------------------------------- Link status -------------------------------*/
514
515/*
Jay Vosburghff59c452006-03-27 13:27:43 -0800516 * Set the carrier state for the master according to the state of its
517 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
518 * do special 802.3ad magic.
519 *
520 * Returns zero if carrier state does not change, nonzero if it does.
521 */
522static int bond_set_carrier(struct bonding *bond)
523{
524 struct slave *slave;
525 int i;
526
527 if (bond->slave_cnt == 0)
528 goto down;
529
530 if (bond->params.mode == BOND_MODE_8023AD)
531 return bond_3ad_set_carrier(bond);
532
533 bond_for_each_slave(bond, slave, i) {
534 if (slave->link == BOND_LINK_UP) {
535 if (!netif_carrier_ok(bond->dev)) {
536 netif_carrier_on(bond->dev);
537 return 1;
538 }
539 return 0;
540 }
541 }
542
543down:
544 if (netif_carrier_ok(bond->dev)) {
545 netif_carrier_off(bond->dev);
546 return 1;
547 }
548 return 0;
549}
550
551/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 * Get link speed and duplex from the slave's base driver
553 * using ethtool. If for some reason the call fails or the
554 * values are invalid, fake speed and duplex to 100/Full
555 * and return error.
556 */
557static int bond_update_speed_duplex(struct slave *slave)
558{
559 struct net_device *slave_dev = slave->dev;
David Decotigny5d305302011-04-13 15:22:31 +0000560 struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
561 u32 slave_speed;
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700562 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564 /* Fake speed and duplex */
565 slave->speed = SPEED_100;
566 slave->duplex = DUPLEX_FULL;
567
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700568 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700571 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
572 if (res < 0)
573 return -1;
574
David Decotigny5d305302011-04-13 15:22:31 +0000575 slave_speed = ethtool_cmd_speed(&etool);
Jiri Pirko6f92c662011-06-01 10:36:33 +0000576 if (slave_speed == 0 || slave_speed == ((__u32) -1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
579 switch (etool.duplex) {
580 case DUPLEX_FULL:
581 case DUPLEX_HALF:
582 break;
583 default:
584 return -1;
585 }
586
David Decotigny5d305302011-04-13 15:22:31 +0000587 slave->speed = slave_speed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 slave->duplex = etool.duplex;
589
590 return 0;
591}
592
593/*
594 * if <dev> supports MII link status reporting, check its link status.
595 *
596 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000597 * depending upon the setting of the use_carrier parameter.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 *
599 * Return either BMSR_LSTATUS, meaning that the link is up (or we
600 * can't tell and just pretend it is), or 0, meaning that the link is
601 * down.
602 *
603 * If reporting is non-zero, instead of faking link up, return -1 if
604 * both ETHTOOL and MII ioctls fail (meaning the device does not
605 * support them). If use_carrier is set, return whatever it says.
606 * It'd be nice if there was a good way to tell if a driver supports
607 * netif_carrier, but there really isn't.
608 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000609static int bond_check_dev_link(struct bonding *bond,
610 struct net_device *slave_dev, int reporting)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800612 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Jiri Bohacd9d52832009-10-28 22:23:54 -0700613 int (*ioctl)(struct net_device *, struct ifreq *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 struct ifreq ifr;
615 struct mii_ioctl_data *mii;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Petri Gynther6c988852009-08-28 12:05:15 +0000617 if (!reporting && !netif_running(slave_dev))
618 return 0;
619
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800620 if (bond->params.use_carrier)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
Jiri Pirko29112f42009-04-24 01:58:23 +0000623 /* Try to get link status using Ethtool first. */
624 if (slave_dev->ethtool_ops) {
625 if (slave_dev->ethtool_ops->get_link) {
626 u32 link;
627
628 link = slave_dev->ethtool_ops->get_link(slave_dev);
629
630 return link ? BMSR_LSTATUS : 0;
631 }
632 }
633
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000634 /* Ethtool can't be used, fallback to MII ioctls. */
Stephen Hemmingereb7cc592008-11-19 21:56:05 -0800635 ioctl = slave_ops->ndo_do_ioctl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 if (ioctl) {
637 /* TODO: set pointer to correct ioctl on a per team member */
638 /* bases to make this more efficient. that is, once */
639 /* we determine the correct ioctl, we will always */
640 /* call it and not the others for that team */
641 /* member. */
642
643 /*
644 * We cannot assume that SIOCGMIIPHY will also read a
645 * register; not all network drivers (e.g., e100)
646 * support that.
647 */
648
649 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
650 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
651 mii = if_mii(&ifr);
652 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
653 mii->reg_num = MII_BMSR;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000654 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
655 return mii->val_out & BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 }
657 }
658
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700659 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 * If reporting, report that either there's no dev->do_ioctl,
Matthew Wilcox61a44b92007-07-31 14:00:02 -0700661 * or both SIOCGMIIREG and get_link failed (meaning that we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 * cannot report link status). If not reporting, pretend
663 * we're ok.
664 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000665 return reporting ? -1 : BMSR_LSTATUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666}
667
668/*----------------------------- Multicast list ------------------------------*/
669
670/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 * Push the promiscuity flag down to appropriate slaves
672 */
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700673static int bond_set_promiscuity(struct bonding *bond, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700675 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 if (USES_PRIMARY(bond->params.mode)) {
677 /* write lock already acquired */
678 if (bond->curr_active_slave) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700679 err = dev_set_promiscuity(bond->curr_active_slave->dev,
680 inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 }
682 } else {
683 struct slave *slave;
684 int i;
685 bond_for_each_slave(bond, slave, i) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700686 err = dev_set_promiscuity(slave->dev, inc);
687 if (err)
688 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 }
690 }
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700691 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692}
693
694/*
695 * Push the allmulti flag down to all slaves
696 */
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700697static int bond_set_allmulti(struct bonding *bond, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700699 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 if (USES_PRIMARY(bond->params.mode)) {
701 /* write lock already acquired */
702 if (bond->curr_active_slave) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700703 err = dev_set_allmulti(bond->curr_active_slave->dev,
704 inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 }
706 } else {
707 struct slave *slave;
708 int i;
709 bond_for_each_slave(bond, slave, i) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700710 err = dev_set_allmulti(slave->dev, inc);
711 if (err)
712 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 }
714 }
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700715 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716}
717
718/*
719 * Add a Multicast address to slaves
720 * according to mode
721 */
Jiri Pirko22bedad32010-04-01 21:22:57 +0000722static void bond_mc_add(struct bonding *bond, void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723{
724 if (USES_PRIMARY(bond->params.mode)) {
725 /* write lock already acquired */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000726 if (bond->curr_active_slave)
Jiri Pirko22bedad32010-04-01 21:22:57 +0000727 dev_mc_add(bond->curr_active_slave->dev, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 } else {
729 struct slave *slave;
730 int i;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000731
732 bond_for_each_slave(bond, slave, i)
Jiri Pirko22bedad32010-04-01 21:22:57 +0000733 dev_mc_add(slave->dev, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 }
735}
736
737/*
738 * Remove a multicast address from slave
739 * according to mode
740 */
Jiri Pirko22bedad32010-04-01 21:22:57 +0000741static void bond_mc_del(struct bonding *bond, void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742{
743 if (USES_PRIMARY(bond->params.mode)) {
744 /* write lock already acquired */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000745 if (bond->curr_active_slave)
Jiri Pirko22bedad32010-04-01 21:22:57 +0000746 dev_mc_del(bond->curr_active_slave->dev, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 } else {
748 struct slave *slave;
749 int i;
750 bond_for_each_slave(bond, slave, i) {
Jiri Pirko22bedad32010-04-01 21:22:57 +0000751 dev_mc_del(slave->dev, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 }
753 }
754}
755
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800756
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000757static void __bond_resend_igmp_join_requests(struct net_device *dev)
758{
759 struct in_device *in_dev;
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000760
761 rcu_read_lock();
762 in_dev = __in_dev_get_rcu(dev);
Eric Dumazet866f3b22010-11-18 09:33:19 -0800763 if (in_dev)
David S. Miller24912422010-11-19 13:13:47 -0800764 ip_mc_rejoin_groups(in_dev);
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000765 rcu_read_unlock();
766}
767
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800768/*
769 * Retrieve the list of registered multicast addresses for the bonding
770 * device and retransmit an IGMP JOIN request to the current active
771 * slave.
772 */
773static void bond_resend_igmp_join_requests(struct bonding *bond)
774{
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000775 struct net_device *vlan_dev;
776 struct vlan_entry *vlan;
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800777
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000778 read_lock(&bond->lock);
779
780 /* rejoin all groups on bond device */
781 __bond_resend_igmp_join_requests(bond->dev);
782
783 /* rejoin all groups on vlan devices */
Jiri Pirkocc0e4072011-07-20 04:54:46 +0000784 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
785 rcu_read_lock();
786 vlan_dev = __vlan_find_dev_deep(bond->dev,
787 vlan->vlan_id);
788 rcu_read_unlock();
789 if (vlan_dev)
790 __bond_resend_igmp_join_requests(vlan_dev);
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800791 }
792
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000793 if (--bond->igmp_retrans > 0)
794 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
795
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000796 read_unlock(&bond->lock);
797}
798
stephen hemminger379b7382010-10-15 11:02:56 +0000799static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000800{
801 struct bonding *bond = container_of(work, struct bonding,
Flavio Leitner94265cf2011-05-25 08:38:58 +0000802 mcast_work.work);
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000803 bond_resend_igmp_join_requests(bond);
Jay Vosburgha816c7c2007-02-28 17:03:37 -0800804}
805
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 * flush all members of flush->mc_list from device dev->mc_list
808 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000809static void bond_mc_list_flush(struct net_device *bond_dev,
810 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
Wang Chen454d7c92008-11-12 23:37:49 -0800812 struct bonding *bond = netdev_priv(bond_dev);
Jiri Pirko22bedad32010-04-01 21:22:57 +0000813 struct netdev_hw_addr *ha;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
Jiri Pirko22bedad32010-04-01 21:22:57 +0000815 netdev_for_each_mc_addr(ha, bond_dev)
816 dev_mc_del(slave_dev, ha->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
818 if (bond->params.mode == BOND_MODE_8023AD) {
819 /* del lacpdu mc addr from mc list */
820 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
821
Jiri Pirko22bedad32010-04-01 21:22:57 +0000822 dev_mc_del(slave_dev, lacpdu_multicast);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 }
824}
825
826/*--------------------------- Active slave change ---------------------------*/
827
828/*
829 * Update the mc list and multicast-related flags for the new and
830 * old active slaves (if any) according to the multicast mode, and
831 * promiscuous flags unconditionally.
832 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000833static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
834 struct slave *old_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835{
Jiri Pirko22bedad32010-04-01 21:22:57 +0000836 struct netdev_hw_addr *ha;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000838 if (!USES_PRIMARY(bond->params.mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 /* nothing to do - mc list is already up-to-date on
840 * all slaves
841 */
842 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
844 if (old_active) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000845 if (bond->dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 dev_set_promiscuity(old_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000848 if (bond->dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 dev_set_allmulti(old_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
Jiri Pirko22bedad32010-04-01 21:22:57 +0000851 netdev_for_each_mc_addr(ha, bond->dev)
852 dev_mc_del(old_active->dev, ha->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 }
854
855 if (new_active) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700856 /* FIXME: Signal errors upstream. */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000857 if (bond->dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 dev_set_promiscuity(new_active->dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000860 if (bond->dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 dev_set_allmulti(new_active->dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
Jiri Pirko22bedad32010-04-01 21:22:57 +0000863 netdev_for_each_mc_addr(ha, bond->dev)
864 dev_mc_add(new_active->dev, ha->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 }
866}
867
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700868/*
869 * bond_do_fail_over_mac
870 *
871 * Perform special MAC address swapping for fail_over_mac settings
872 *
873 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
874 */
875static void bond_do_fail_over_mac(struct bonding *bond,
876 struct slave *new_active,
877 struct slave *old_active)
Hannes Eder1f78d9f2009-02-14 11:15:33 +0000878 __releases(&bond->curr_slave_lock)
879 __releases(&bond->lock)
880 __acquires(&bond->lock)
881 __acquires(&bond->curr_slave_lock)
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700882{
883 u8 tmp_mac[ETH_ALEN];
884 struct sockaddr saddr;
885 int rv;
886
887 switch (bond->params.fail_over_mac) {
888 case BOND_FOM_ACTIVE:
889 if (new_active)
890 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
891 new_active->dev->addr_len);
892 break;
893 case BOND_FOM_FOLLOW:
894 /*
895 * if new_active && old_active, swap them
896 * if just old_active, do nothing (going to no active slave)
897 * if just new_active, set new_active to bond's MAC
898 */
899 if (!new_active)
900 return;
901
902 write_unlock_bh(&bond->curr_slave_lock);
903 read_unlock(&bond->lock);
904
905 if (old_active) {
906 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
907 memcpy(saddr.sa_data, old_active->dev->dev_addr,
908 ETH_ALEN);
909 saddr.sa_family = new_active->dev->type;
910 } else {
911 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
912 saddr.sa_family = bond->dev->type;
913 }
914
915 rv = dev_set_mac_address(new_active->dev, &saddr);
916 if (rv) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800917 pr_err("%s: Error %d setting MAC of slave %s\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700918 bond->dev->name, -rv, new_active->dev->name);
919 goto out;
920 }
921
922 if (!old_active)
923 goto out;
924
925 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
926 saddr.sa_family = old_active->dev->type;
927
928 rv = dev_set_mac_address(old_active->dev, &saddr);
929 if (rv)
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800930 pr_err("%s: Error %d setting MAC of slave %s\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700931 bond->dev->name, -rv, new_active->dev->name);
932out:
933 read_lock(&bond->lock);
934 write_lock_bh(&bond->curr_slave_lock);
935 break;
936 default:
Joe Perchesa4aee5c2009-12-13 20:06:07 -0800937 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700938 bond->dev->name, bond->params.fail_over_mac);
939 break;
940 }
941
942}
943
Jiri Pirkoa5499522009-09-25 03:28:09 +0000944static bool bond_should_change_active(struct bonding *bond)
945{
946 struct slave *prim = bond->primary_slave;
947 struct slave *curr = bond->curr_active_slave;
948
949 if (!prim || !curr || curr->link != BOND_LINK_UP)
950 return true;
951 if (bond->force_primary) {
952 bond->force_primary = false;
953 return true;
954 }
955 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
956 (prim->speed < curr->speed ||
957 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
958 return false;
959 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
960 return false;
961 return true;
962}
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700963
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964/**
965 * find_best_interface - select the best available slave to be the active one
966 * @bond: our bonding struct
967 *
968 * Warning: Caller must hold curr_slave_lock for writing.
969 */
970static struct slave *bond_find_best_slave(struct bonding *bond)
971{
972 struct slave *new_active, *old_active;
973 struct slave *bestslave = NULL;
974 int mintime = bond->params.updelay;
975 int i;
976
Nicolas de Pesloüan49b4ad92009-10-07 14:11:00 -0700977 new_active = bond->curr_active_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
979 if (!new_active) { /* there were no active slaves left */
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000980 if (bond->slave_cnt > 0) /* found one slave */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 new_active = bond->first_slave;
Stephen Hemminger3d632c32009-06-12 19:02:48 +0000982 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 return NULL; /* still no slave, return NULL */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 }
985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 if ((bond->primary_slave) &&
Jiri Pirkoa5499522009-09-25 03:28:09 +0000987 bond->primary_slave->link == BOND_LINK_UP &&
988 bond_should_change_active(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 new_active = bond->primary_slave;
990 }
991
992 /* remember where to stop iterating over the slaves */
993 old_active = new_active;
994
995 bond_for_each_slave_from(bond, new_active, i, old_active) {
Jiri Pirkob9f60252009-08-31 11:09:38 +0000996 if (new_active->link == BOND_LINK_UP) {
997 return new_active;
998 } else if (new_active->link == BOND_LINK_BACK &&
999 IS_UP(new_active->dev)) {
1000 /* link up, but waiting for stabilization */
1001 if (new_active->delay < mintime) {
1002 mintime = new_active->delay;
1003 bestslave = new_active;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
1005 }
1006 }
1007
1008 return bestslave;
1009}
1010
Ben Hutchingsad246c92011-04-26 15:25:52 +00001011static bool bond_should_notify_peers(struct bonding *bond)
1012{
1013 struct slave *slave = bond->curr_active_slave;
1014
1015 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1016 bond->dev->name, slave ? slave->dev->name : "NULL");
1017
1018 if (!slave || !bond->send_peer_notif ||
1019 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1020 return false;
1021
1022 bond->send_peer_notif--;
1023 return true;
1024}
1025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026/**
1027 * change_active_interface - change the active slave into the specified one
1028 * @bond: our bonding struct
1029 * @new: the new slave to make the active one
1030 *
1031 * Set the new slave to the bond's settings and unset them on the old
1032 * curr_active_slave.
1033 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1034 *
1035 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1036 * because it is apparently the best available slave we have, even though its
1037 * updelay hasn't timed out yet.
1038 *
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001039 * If new_active is not NULL, caller must hold bond->lock for read and
1040 * curr_slave_lock for write_bh.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001042void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043{
1044 struct slave *old_active = bond->curr_active_slave;
1045
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001046 if (old_active == new_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
1049 if (new_active) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07001050 new_active->jiffies = jiffies;
1051
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 if (new_active->link == BOND_LINK_BACK) {
1053 if (USES_PRIMARY(bond->params.mode)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001054 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1055 bond->dev->name, new_active->dev->name,
1056 (bond->params.updelay - new_active->delay) * bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 }
1058
1059 new_active->delay = 0;
1060 new_active->link = BOND_LINK_UP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001062 if (bond->params.mode == BOND_MODE_8023AD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Holger Eitzenberger58402052008-12-09 23:07:13 -08001065 if (bond_is_lb(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 } else {
1068 if (USES_PRIMARY(bond->params.mode)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001069 pr_info("%s: making interface %s the new active one.\n",
1070 bond->dev->name, new_active->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 }
1072 }
1073 }
1074
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001075 if (USES_PRIMARY(bond->params.mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 bond_mc_swap(bond, new_active, old_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
Holger Eitzenberger58402052008-12-09 23:07:13 -08001078 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 bond_alb_handle_active_change(bond, new_active);
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001080 if (old_active)
1081 bond_set_slave_inactive_flags(old_active);
1082 if (new_active)
1083 bond_set_slave_active_flags(new_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 } else {
1085 bond->curr_active_slave = new_active;
1086 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001087
1088 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001089 if (old_active)
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001090 bond_set_slave_inactive_flags(old_active);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001091
1092 if (new_active) {
Ben Hutchingsad246c92011-04-26 15:25:52 +00001093 bool should_notify_peers = false;
1094
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001095 bond_set_slave_active_flags(new_active);
Moni Shoua2ab82852007-10-09 19:43:39 -07001096
Or Gerlitz709f8a42008-06-13 18:12:01 -07001097 if (bond->params.fail_over_mac)
1098 bond_do_fail_over_mac(bond, new_active,
1099 old_active);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001100
Ben Hutchingsad246c92011-04-26 15:25:52 +00001101 if (netif_running(bond->dev)) {
1102 bond->send_peer_notif =
1103 bond->params.num_peer_notif;
1104 should_notify_peers =
1105 bond_should_notify_peers(bond);
1106 }
1107
Or Gerlitz01f31092008-06-13 18:12:02 -07001108 write_unlock_bh(&bond->curr_slave_lock);
1109 read_unlock(&bond->lock);
1110
Moni Shoua75c78502009-09-15 02:37:40 -07001111 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
Ben Hutchingsad246c92011-04-26 15:25:52 +00001112 if (should_notify_peers)
1113 netdev_bonding_change(bond->dev,
1114 NETDEV_NOTIFY_PEERS);
Or Gerlitz01f31092008-06-13 18:12:02 -07001115
1116 read_lock(&bond->lock);
1117 write_lock_bh(&bond->curr_slave_lock);
Moni Shoua7893b242008-05-17 21:10:12 -07001118 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001119 }
Andy Gospodareka2fd9402010-03-25 14:49:05 +00001120
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00001121 /* resend IGMP joins since active slave has changed or
Flavio Leitner94265cf2011-05-25 08:38:58 +00001122 * all were sent on curr_active_slave.
1123 * resend only if bond is brought up with the affected
1124 * bonding modes and the retransmission is enabled */
1125 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1126 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1127 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
Flavio Leitnerc2952c32010-10-05 14:23:59 +00001128 bond->igmp_retrans = bond->params.resend_igmp;
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00001129 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
Andy Gospodareka2fd9402010-03-25 14:49:05 +00001130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131}
1132
1133/**
1134 * bond_select_active_slave - select a new active slave, if needed
1135 * @bond: our bonding struct
1136 *
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001137 * This functions should be called when one of the following occurs:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 * - The old curr_active_slave has been released or lost its link.
1139 * - The primary_slave has got its link back.
1140 * - A slave has got its link back and there's no old curr_active_slave.
1141 *
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001142 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001144void bond_select_active_slave(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145{
1146 struct slave *best_slave;
Jay Vosburghff59c452006-03-27 13:27:43 -08001147 int rv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
1149 best_slave = bond_find_best_slave(bond);
1150 if (best_slave != bond->curr_active_slave) {
1151 bond_change_active_slave(bond, best_slave);
Jay Vosburghff59c452006-03-27 13:27:43 -08001152 rv = bond_set_carrier(bond);
1153 if (!rv)
1154 return;
1155
1156 if (netif_carrier_ok(bond->dev)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001157 pr_info("%s: first active interface up!\n",
1158 bond->dev->name);
Jay Vosburghff59c452006-03-27 13:27:43 -08001159 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001160 pr_info("%s: now running without any active interface !\n",
1161 bond->dev->name);
Jay Vosburghff59c452006-03-27 13:27:43 -08001162 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 }
1164}
1165
1166/*--------------------------- slave list handling ---------------------------*/
1167
1168/*
1169 * This function attaches the slave to the end of list.
1170 *
1171 * bond->lock held for writing by caller.
1172 */
1173static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1174{
1175 if (bond->first_slave == NULL) { /* attaching the first slave */
1176 new_slave->next = new_slave;
1177 new_slave->prev = new_slave;
1178 bond->first_slave = new_slave;
1179 } else {
1180 new_slave->next = bond->first_slave;
1181 new_slave->prev = bond->first_slave->prev;
1182 new_slave->next->prev = new_slave;
1183 new_slave->prev->next = new_slave;
1184 }
1185
1186 bond->slave_cnt++;
1187}
1188
1189/*
1190 * This function detaches the slave from the list.
1191 * WARNING: no check is made to verify if the slave effectively
1192 * belongs to <bond>.
1193 * Nothing is freed on return, structures are just unchained.
1194 * If any slave pointer in bond was pointing to <slave>,
1195 * it should be changed by the calling function.
1196 *
1197 * bond->lock held for writing by caller.
1198 */
1199static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1200{
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001201 if (slave->next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 slave->next->prev = slave->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001204 if (slave->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 slave->prev->next = slave->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
1207 if (bond->first_slave == slave) { /* slave is the first slave */
1208 if (bond->slave_cnt > 1) { /* there are more slave */
1209 bond->first_slave = slave->next;
1210 } else {
1211 bond->first_slave = NULL; /* slave was the last one */
1212 }
1213 }
1214
1215 slave->next = NULL;
1216 slave->prev = NULL;
1217 bond->slave_cnt--;
1218}
1219
WANG Congf6dc31a2010-05-06 00:48:51 -07001220#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001221static inline int slave_enable_netpoll(struct slave *slave)
WANG Congf6dc31a2010-05-06 00:48:51 -07001222{
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001223 struct netpoll *np;
1224 int err = 0;
WANG Congf6dc31a2010-05-06 00:48:51 -07001225
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001226 np = kzalloc(sizeof(*np), GFP_KERNEL);
1227 err = -ENOMEM;
1228 if (!np)
1229 goto out;
1230
1231 np->dev = slave->dev;
WANG Congcefa9992011-06-19 16:13:01 -07001232 strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001233 err = __netpoll_setup(np);
1234 if (err) {
1235 kfree(np);
1236 goto out;
WANG Congf6dc31a2010-05-06 00:48:51 -07001237 }
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001238 slave->np = np;
1239out:
1240 return err;
1241}
1242static inline void slave_disable_netpoll(struct slave *slave)
1243{
1244 struct netpoll *np = slave->np;
1245
1246 if (!np)
1247 return;
1248
1249 slave->np = NULL;
1250 synchronize_rcu_bh();
1251 __netpoll_cleanup(np);
1252 kfree(np);
1253}
1254static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1255{
1256 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1257 return false;
1258 if (!slave_dev->netdev_ops->ndo_poll_controller)
1259 return false;
1260 return true;
WANG Congf6dc31a2010-05-06 00:48:51 -07001261}
1262
1263static void bond_poll_controller(struct net_device *bond_dev)
1264{
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001265}
1266
1267static void __bond_netpoll_cleanup(struct bonding *bond)
1268{
Neil Hormanc2355e12010-10-13 16:01:49 +00001269 struct slave *slave;
1270 int i;
1271
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001272 bond_for_each_slave(bond, slave, i)
1273 if (IS_UP(slave->dev))
1274 slave_disable_netpoll(slave);
WANG Congf6dc31a2010-05-06 00:48:51 -07001275}
WANG Congf6dc31a2010-05-06 00:48:51 -07001276static void bond_netpoll_cleanup(struct net_device *bond_dev)
1277{
1278 struct bonding *bond = netdev_priv(bond_dev);
WANG Congf6dc31a2010-05-06 00:48:51 -07001279
1280 read_lock(&bond->lock);
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001281 __bond_netpoll_cleanup(bond);
WANG Congf6dc31a2010-05-06 00:48:51 -07001282 read_unlock(&bond->lock);
1283}
1284
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001285static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1286{
1287 struct bonding *bond = netdev_priv(dev);
1288 struct slave *slave;
1289 int i, err = 0;
WANG Congf6dc31a2010-05-06 00:48:51 -07001290
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001291 read_lock(&bond->lock);
1292 bond_for_each_slave(bond, slave, i) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001293 err = slave_enable_netpoll(slave);
1294 if (err) {
1295 __bond_netpoll_cleanup(bond);
1296 break;
1297 }
1298 }
1299 read_unlock(&bond->lock);
1300 return err;
1301}
1302
1303static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1304{
1305 return bond->dev->npinfo;
1306}
1307
1308#else
1309static inline int slave_enable_netpoll(struct slave *slave)
1310{
1311 return 0;
1312}
1313static inline void slave_disable_netpoll(struct slave *slave)
1314{
1315}
WANG Congf6dc31a2010-05-06 00:48:51 -07001316static void bond_netpoll_cleanup(struct net_device *bond_dev)
1317{
1318}
WANG Congf6dc31a2010-05-06 00:48:51 -07001319#endif
1320
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321/*---------------------------------- IOCTL ----------------------------------*/
1322
Adrian Bunk4ad072c2007-07-09 11:51:12 -07001323static int bond_sethwaddr(struct net_device *bond_dev,
1324 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001326 pr_debug("bond_dev=%p\n", bond_dev);
1327 pr_debug("slave_dev=%p\n", slave_dev);
1328 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1330 return 0;
1331}
1332
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001333static u32 bond_fix_features(struct net_device *dev, u32 features)
1334{
1335 struct slave *slave;
1336 struct bonding *bond = netdev_priv(dev);
1337 u32 mask;
1338 int i;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001339
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001340 read_lock(&bond->lock);
1341
1342 if (!bond->first_slave) {
1343 /* Disable adding VLANs to empty bond. But why? --mq */
1344 features |= NETIF_F_VLAN_CHALLENGED;
1345 goto out;
1346 }
1347
1348 mask = features;
1349 features &= ~NETIF_F_ONE_FOR_ALL;
1350 features |= NETIF_F_ALL_FOR_ALL;
1351
1352 bond_for_each_slave(bond, slave, i) {
1353 features = netdev_increment_features(features,
1354 slave->dev->features,
1355 mask);
1356 }
1357
1358out:
1359 read_unlock(&bond->lock);
1360 return features;
1361}
1362
Michał Mirosław62f2a3a2011-07-13 14:10:29 +00001363#define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1364 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1365 NETIF_F_HIGHDMA | NETIF_F_LRO)
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001366
1367static void bond_compute_features(struct bonding *bond)
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001368{
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001369 struct slave *slave;
1370 struct net_device *bond_dev = bond->dev;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001371 u32 vlan_features = BOND_VLAN_FEATURES;
1372 unsigned short max_hard_header_len = ETH_HLEN;
Jay Vosburgh8e3babc2005-11-04 18:45:45 -08001373 int i;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001374
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001375 read_lock(&bond->lock);
Herbert Xub63365a2008-10-23 01:11:29 -07001376
1377 if (!bond->first_slave)
1378 goto done;
1379
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001380 bond_for_each_slave(bond, slave, i) {
Jay Vosburgh278339a2009-08-28 12:05:12 +00001381 vlan_features = netdev_increment_features(vlan_features,
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001382 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1383
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001384 if (slave->dev->hard_header_len > max_hard_header_len)
1385 max_hard_header_len = slave->dev->hard_header_len;
1386 }
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001387
Herbert Xub63365a2008-10-23 01:11:29 -07001388done:
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001389 bond_dev->vlan_features = vlan_features;
Jay Vosburgh54ef3132006-09-22 21:53:39 -07001390 bond_dev->hard_header_len = max_hard_header_len;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001391
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001392 read_unlock(&bond->lock);
1393
1394 netdev_change_features(bond_dev);
Arthur Kepner8531c5f2005-08-23 01:34:53 -04001395}
1396
Moni Shoua872254d2007-10-09 19:43:38 -07001397static void bond_setup_by_slave(struct net_device *bond_dev,
1398 struct net_device *slave_dev)
1399{
Wang Chen454d7c92008-11-12 23:37:49 -08001400 struct bonding *bond = netdev_priv(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07001401
Stephen Hemminger00829822008-11-20 20:14:53 -08001402 bond_dev->header_ops = slave_dev->header_ops;
Moni Shoua872254d2007-10-09 19:43:38 -07001403
1404 bond_dev->type = slave_dev->type;
1405 bond_dev->hard_header_len = slave_dev->hard_header_len;
1406 bond_dev->addr_len = slave_dev->addr_len;
1407
1408 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1409 slave_dev->addr_len);
Moni Shouad90a1622007-10-09 19:43:43 -07001410 bond->setup_by_slave = 1;
Moni Shoua872254d2007-10-09 19:43:38 -07001411}
1412
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001413/* On bonding slaves other than the currently active slave, suppress
Jiri Pirko3aba8912011-04-19 03:48:16 +00001414 * duplicates except for alb non-mcast/bcast.
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001415 */
1416static bool bond_should_deliver_exact_match(struct sk_buff *skb,
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001417 struct slave *slave,
1418 struct bonding *bond)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001419{
Jiri Pirko2d7011c2011-03-16 08:46:43 +00001420 if (bond_is_slave_inactive(slave)) {
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001421 if (bond->params.mode == BOND_MODE_ALB &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001422 skb->pkt_type != PACKET_BROADCAST &&
1423 skb->pkt_type != PACKET_MULTICAST)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001424 return false;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001425 return true;
1426 }
1427 return false;
1428}
1429
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001430static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001431{
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001432 struct sk_buff *skb = *pskb;
Jiri Pirkof1c17752011-03-12 03:14:35 +00001433 struct slave *slave;
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001434 struct bonding *bond;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001435
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001436 skb = skb_share_check(skb, GFP_ATOMIC);
1437 if (unlikely(!skb))
1438 return RX_HANDLER_CONSUMED;
1439
1440 *pskb = skb;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001441
Jiri Pirko35d48902011-03-22 02:38:12 +00001442 slave = bond_slave_get_rcu(skb->dev);
1443 bond = slave->bond;
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001444
1445 if (bond->params.arp_interval)
Jiri Pirkof1c17752011-03-12 03:14:35 +00001446 slave->dev->last_rx = jiffies;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001447
Jiri Pirko3aba8912011-04-19 03:48:16 +00001448 if (bond->recv_probe) {
1449 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1450
1451 if (likely(nskb)) {
1452 bond->recv_probe(nskb, bond, slave);
1453 dev_kfree_skb(nskb);
1454 }
1455 }
1456
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001457 if (bond_should_deliver_exact_match(skb, slave, bond)) {
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001458 return RX_HANDLER_EXACT;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001459 }
1460
Jiri Pirko35d48902011-03-22 02:38:12 +00001461 skb->dev = bond->dev;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001462
Jiri Pirko0bd80da2011-03-16 08:45:23 +00001463 if (bond->params.mode == BOND_MODE_ALB &&
Jiri Pirko35d48902011-03-22 02:38:12 +00001464 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001465 skb->pkt_type == PACKET_HOST) {
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001466
Changli Gao541ac7c2011-03-02 21:07:14 +00001467 if (unlikely(skb_cow_head(skb,
1468 skb->data - skb_mac_header(skb)))) {
1469 kfree_skb(skb);
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001470 return RX_HANDLER_CONSUMED;
Changli Gao541ac7c2011-03-02 21:07:14 +00001471 }
Jiri Pirko35d48902011-03-22 02:38:12 +00001472 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001473 }
1474
Jiri Pirko8a4eb572011-03-12 03:14:39 +00001475 return RX_HANDLER_ANOTHER;
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001476}
1477
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478/* enslave device <slave> to bond device <master> */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001479int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480{
Wang Chen454d7c92008-11-12 23:37:49 -08001481 struct bonding *bond = netdev_priv(bond_dev);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001482 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 struct slave *new_slave = NULL;
Jiri Pirko22bedad32010-04-01 21:22:57 +00001484 struct netdev_hw_addr *ha;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 struct sockaddr addr;
1486 int link_reporting;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 int res = 0;
1488
nsxfreddy@gmail.com552709d2005-09-21 14:18:04 -05001489 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001490 slave_ops->ndo_do_ioctl == NULL) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001491 pr_warning("%s: Warning: no link monitoring support for %s\n",
1492 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 }
1494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 /* already enslaved */
1496 if (slave_dev->flags & IFF_SLAVE) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001497 pr_debug("Error, Device was already enslaved\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 return -EBUSY;
1499 }
1500
1501 /* vlan challenged mutual exclusion */
1502 /* no need to lock since we're protected by rtnl_lock */
1503 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001504 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
Jiri Pirkocc0e4072011-07-20 04:54:46 +00001505 if (bond_vlan_used(bond)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001506 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1507 bond_dev->name, slave_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 return -EPERM;
1509 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001510 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1511 bond_dev->name, slave_dev->name,
1512 slave_dev->name, bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 }
1514 } else {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001515 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 }
1517
Jay Vosburgh217df672005-09-26 16:11:50 -07001518 /*
1519 * Old ifenslave binaries are no longer supported. These can
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001520 * be identified with moderate accuracy by the state of the slave:
Jay Vosburgh217df672005-09-26 16:11:50 -07001521 * the current ifenslave will set the interface down prior to
1522 * enslaving it; the old ifenslave will not.
1523 */
1524 if ((slave_dev->flags & IFF_UP)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001525 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
Jay Vosburgh217df672005-09-26 16:11:50 -07001526 slave_dev->name);
1527 res = -EPERM;
1528 goto err_undo_flags;
1529 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
Moni Shoua872254d2007-10-09 19:43:38 -07001531 /* set bonding device ether type by slave - bonding netdevices are
1532 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1533 * there is a need to override some of the type dependent attribs/funcs.
1534 *
1535 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1536 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1537 */
1538 if (bond->slave_cnt == 0) {
Moni Shouae36b9d12009-07-15 04:56:31 +00001539 if (bond_dev->type != slave_dev->type) {
Moni Shouae36b9d12009-07-15 04:56:31 +00001540 pr_debug("%s: change device type from %d to %d\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001541 bond_dev->name,
1542 bond_dev->type, slave_dev->type);
Moni Shoua75c78502009-09-15 02:37:40 -07001543
Jiri Pirko3ca5b402010-03-10 10:29:35 +00001544 res = netdev_bonding_change(bond_dev,
1545 NETDEV_PRE_TYPE_CHANGE);
1546 res = notifier_to_errno(res);
1547 if (res) {
1548 pr_err("%s: refused to change device type\n",
1549 bond_dev->name);
1550 res = -EBUSY;
1551 goto err_undo_flags;
1552 }
Moni Shoua75c78502009-09-15 02:37:40 -07001553
Jiri Pirko32a806c2010-03-19 04:00:23 +00001554 /* Flush unicast and multicast addresses */
Jiri Pirkoa748ee22010-04-01 21:22:09 +00001555 dev_uc_flush(bond_dev);
Jiri Pirko22bedad32010-04-01 21:22:57 +00001556 dev_mc_flush(bond_dev);
Jiri Pirko32a806c2010-03-19 04:00:23 +00001557
Moni Shouae36b9d12009-07-15 04:56:31 +00001558 if (slave_dev->type != ARPHRD_ETHER)
1559 bond_setup_by_slave(bond_dev, slave_dev);
Neil Horman550fd082011-07-26 06:05:38 +00001560 else {
Moni Shouae36b9d12009-07-15 04:56:31 +00001561 ether_setup(bond_dev);
Neil Horman550fd082011-07-26 06:05:38 +00001562 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1563 }
Moni Shoua75c78502009-09-15 02:37:40 -07001564
Jiri Pirko93d9b7d2010-03-10 10:28:56 +00001565 netdev_bonding_change(bond_dev,
1566 NETDEV_POST_TYPE_CHANGE);
Moni Shouae36b9d12009-07-15 04:56:31 +00001567 }
Moni Shoua872254d2007-10-09 19:43:38 -07001568 } else if (bond_dev->type != slave_dev->type) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001569 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1570 slave_dev->name,
1571 slave_dev->type, bond_dev->type);
1572 res = -EINVAL;
1573 goto err_undo_flags;
Moni Shoua872254d2007-10-09 19:43:38 -07001574 }
1575
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001576 if (slave_ops->ndo_set_mac_address == NULL) {
Moni Shoua2ab82852007-10-09 19:43:39 -07001577 if (bond->slave_cnt == 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001578 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1579 bond_dev->name);
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001580 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1581 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001582 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1583 bond_dev->name);
Moni Shoua2ab82852007-10-09 19:43:39 -07001584 res = -EOPNOTSUPP;
1585 goto err_undo_flags;
1586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 }
1588
Amerigo Wang8d8fc292011-05-19 21:39:10 +00001589 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1590
Jiri Pirkoc20811a2010-05-19 01:14:29 +00001591 /* If this is the first slave, then we need to set the master's hardware
1592 * address to be the same as the slave's. */
David Strandd13a2cb2010-12-01 11:43:08 -08001593 if (is_zero_ether_addr(bond->dev->dev_addr))
Jiri Pirkoc20811a2010-05-19 01:14:29 +00001594 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1595 slave_dev->addr_len);
1596
1597
Joe Jin243cb4e2007-02-06 14:16:40 -08001598 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 if (!new_slave) {
1600 res = -ENOMEM;
1601 goto err_undo_flags;
1602 }
1603
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00001604 /*
1605 * Set the new_slave's queue_id to be zero. Queue ID mapping
1606 * is set via sysfs or module option if desired.
1607 */
1608 new_slave->queue_id = 0;
1609
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001610 /* Save slave's original mtu and then set it to match the bond */
1611 new_slave->original_mtu = slave_dev->mtu;
1612 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1613 if (res) {
1614 pr_debug("Error %d calling dev_set_mtu\n", res);
1615 goto err_free;
1616 }
1617
Jay Vosburgh217df672005-09-26 16:11:50 -07001618 /*
1619 * Save slave's original ("permanent") mac address for modes
1620 * that need it, and for restoring it upon release, and then
1621 * set it to the master's address
1622 */
1623 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
Jay Vosburghdd957c52007-10-09 19:57:24 -07001625 if (!bond->params.fail_over_mac) {
Moni Shoua2ab82852007-10-09 19:43:39 -07001626 /*
1627 * Set slave to master's mac address. The application already
1628 * set the master's mac address to that of the first slave
1629 */
1630 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1631 addr.sa_family = slave_dev->type;
1632 res = dev_set_mac_address(slave_dev, &addr);
1633 if (res) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001634 pr_debug("Error %d calling set_mac_address\n", res);
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001635 goto err_restore_mtu;
Moni Shoua2ab82852007-10-09 19:43:39 -07001636 }
Jay Vosburgh217df672005-09-26 16:11:50 -07001637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638
Jiri Pirko1765a572011-02-12 06:48:36 +00001639 res = netdev_set_bond_master(slave_dev, bond_dev);
Jay Vosburghc2edacf2007-07-09 10:42:47 -07001640 if (res) {
Jiri Pirko1765a572011-02-12 06:48:36 +00001641 pr_debug("Error %d calling netdev_set_bond_master\n", res);
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001642 goto err_restore_mac;
Jay Vosburghc2edacf2007-07-09 10:42:47 -07001643 }
Jiri Pirko5b2c4dd2011-02-23 09:05:42 +00001644
Jay Vosburgh217df672005-09-26 16:11:50 -07001645 /* open the slave since the application closed it */
1646 res = dev_open(slave_dev);
1647 if (res) {
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001648 pr_debug("Opening slave %s failed\n", slave_dev->name);
Jiri Pirko35d48902011-03-22 02:38:12 +00001649 goto err_unset_master;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 }
1651
Jiri Pirko35d48902011-03-22 02:38:12 +00001652 new_slave->bond = bond;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 new_slave->dev = slave_dev;
Jay Vosburgh0b680e72006-09-22 21:54:10 -07001654 slave_dev->priv_flags |= IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Holger Eitzenberger58402052008-12-09 23:07:13 -08001656 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 /* bond_alb_init_slave() must be called before all other stages since
1658 * it might fail and we do not want to have to undo everything
1659 */
1660 res = bond_alb_init_slave(bond, new_slave);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001661 if (res)
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001662 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 }
1664
1665 /* If the mode USES_PRIMARY, then the new slave gets the
1666 * master's promisc (and mc) settings only if it becomes the
1667 * curr_active_slave, and that is taken care of later when calling
1668 * bond_change_active()
1669 */
1670 if (!USES_PRIMARY(bond->params.mode)) {
1671 /* set promiscuity level to new slave */
1672 if (bond_dev->flags & IFF_PROMISC) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -07001673 res = dev_set_promiscuity(slave_dev, 1);
1674 if (res)
1675 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 }
1677
1678 /* set allmulti level to new slave */
1679 if (bond_dev->flags & IFF_ALLMULTI) {
Wang Chen7e1a1ac2008-07-14 20:51:36 -07001680 res = dev_set_allmulti(slave_dev, 1);
1681 if (res)
1682 goto err_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 }
1684
David S. Millerb9e40852008-07-15 00:15:08 -07001685 netif_addr_lock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 /* upload master's mc_list to new slave */
Jiri Pirko22bedad32010-04-01 21:22:57 +00001687 netdev_for_each_mc_addr(ha, bond_dev)
1688 dev_mc_add(slave_dev, ha->addr);
David S. Millerb9e40852008-07-15 00:15:08 -07001689 netif_addr_unlock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 }
1691
1692 if (bond->params.mode == BOND_MODE_8023AD) {
1693 /* add lacpdu mc addr to mc list */
1694 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1695
Jiri Pirko22bedad32010-04-01 21:22:57 +00001696 dev_mc_add(slave_dev, lacpdu_multicast);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 }
1698
1699 bond_add_vlans_on_slave(bond, slave_dev);
1700
1701 write_lock_bh(&bond->lock);
1702
1703 bond_attach_slave(bond, new_slave);
1704
1705 new_slave->delay = 0;
1706 new_slave->link_failure_count = 0;
1707
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001708 write_unlock_bh(&bond->lock);
1709
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001710 bond_compute_features(bond);
1711
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001712 read_lock(&bond->lock);
1713
Jay Vosburghf5b2b962006-09-22 21:54:53 -07001714 new_slave->last_arp_rx = jiffies;
1715
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 if (bond->params.miimon && !bond->params.use_carrier) {
1717 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1718
1719 if ((link_reporting == -1) && !bond->params.arp_interval) {
1720 /*
1721 * miimon is set but a bonded network driver
1722 * does not support ETHTOOL/MII and
1723 * arp_interval is not set. Note: if
1724 * use_carrier is enabled, we will never go
1725 * here (because netif_carrier is always
1726 * supported); thus, we don't need to change
1727 * the messages for netif_carrier.
1728 */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001729 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08001730 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 } else if (link_reporting == -1) {
1732 /* unable get link status using mii/ethtool */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001733 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1734 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 }
1736 }
1737
1738 /* check for initial state */
1739 if (!bond->params.miimon ||
1740 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1741 if (bond->params.updelay) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001742 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 new_slave->link = BOND_LINK_BACK;
1744 new_slave->delay = bond->params.updelay;
1745 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001746 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 new_slave->link = BOND_LINK_UP;
1748 }
1749 new_slave->jiffies = jiffies;
1750 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001751 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 new_slave->link = BOND_LINK_DOWN;
1753 }
1754
1755 if (bond_update_speed_duplex(new_slave) &&
1756 (new_slave->link != BOND_LINK_DOWN)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001757 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1758 bond_dev->name, new_slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
1760 if (bond->params.mode == BOND_MODE_8023AD) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001761 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1762 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 }
1764 }
1765
1766 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1767 /* if there is a primary slave, remember it */
Jiri Pirkoa5499522009-09-25 03:28:09 +00001768 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 bond->primary_slave = new_slave;
Jiri Pirkoa5499522009-09-25 03:28:09 +00001770 bond->force_primary = true;
1771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 }
1773
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001774 write_lock_bh(&bond->curr_slave_lock);
1775
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 switch (bond->params.mode) {
1777 case BOND_MODE_ACTIVEBACKUP:
Jay Vosburgh8a8e4472006-09-22 21:56:15 -07001778 bond_set_slave_inactive_flags(new_slave);
1779 bond_select_active_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 break;
1781 case BOND_MODE_8023AD:
1782 /* in 802.3ad mode, the internal mechanism
1783 * will activate the slaves in the selected
1784 * aggregator
1785 */
1786 bond_set_slave_inactive_flags(new_slave);
1787 /* if this is the first slave */
1788 if (bond->slave_cnt == 1) {
1789 SLAVE_AD_INFO(new_slave).id = 1;
1790 /* Initialize AD with the number of times that the AD timer is called in 1 second
1791 * can be called only after the mac address of the bond is set
1792 */
Peter Pan(潘卫平)56d00c672011-06-08 21:19:02 +00001793 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 } else {
1795 SLAVE_AD_INFO(new_slave).id =
1796 SLAVE_AD_INFO(new_slave->prev).id + 1;
1797 }
1798
1799 bond_3ad_bind_slave(new_slave);
1800 break;
1801 case BOND_MODE_TLB:
1802 case BOND_MODE_ALB:
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001803 bond_set_active_slave(new_slave);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001804 bond_set_slave_inactive_flags(new_slave);
Jiri Pirko5a29f782009-03-25 17:23:38 -07001805 bond_select_active_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 break;
1807 default:
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08001808 pr_debug("This slave is always active in trunk mode\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809
1810 /* always active in trunk mode */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001811 bond_set_active_slave(new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
1813 /* In trunking mode there is little meaning to curr_active_slave
1814 * anyway (it holds no special properties of the bond device),
1815 * so we can change it without calling change_active_interface()
1816 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001817 if (!bond->curr_active_slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 bond->curr_active_slave = new_slave;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001819
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 break;
1821 } /* switch(bond_mode) */
1822
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001823 write_unlock_bh(&bond->curr_slave_lock);
1824
Jay Vosburghff59c452006-03-27 13:27:43 -08001825 bond_set_carrier(bond);
1826
WANG Congf6dc31a2010-05-06 00:48:51 -07001827#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001828 slave_dev->npinfo = bond_netpoll_info(bond);
1829 if (slave_dev->npinfo) {
1830 if (slave_enable_netpoll(new_slave)) {
1831 read_unlock(&bond->lock);
1832 pr_info("Error, %s: master_dev is using netpoll, "
1833 "but new slave device does not support netpoll.\n",
1834 bond_dev->name);
1835 res = -EBUSY;
1836 goto err_close;
1837 }
WANG Congf6dc31a2010-05-06 00:48:51 -07001838 }
1839#endif
Amerigo Wang8a8efa22011-02-17 23:43:32 +00001840
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001841 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
Mitch Williamsb76cdba2005-11-09 10:36:41 -08001843 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1844 if (res)
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001845 goto err_close;
Mitch Williamsb76cdba2005-11-09 10:36:41 -08001846
Jiri Pirko35d48902011-03-22 02:38:12 +00001847 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1848 new_slave);
1849 if (res) {
1850 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1851 goto err_dest_symlinks;
1852 }
1853
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001854 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1855 bond_dev->name, slave_dev->name,
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001856 bond_is_active_slave(new_slave) ? "n active" : " backup",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001857 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859 /* enslave is successful */
1860 return 0;
1861
1862/* Undo stages on error */
Jiri Pirko35d48902011-03-22 02:38:12 +00001863err_dest_symlinks:
1864 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866err_close:
1867 dev_close(slave_dev);
1868
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001869err_unset_master:
Jiri Pirko1765a572011-02-12 06:48:36 +00001870 netdev_set_bond_master(slave_dev, NULL);
Jay Vosburgh569f0c42008-05-02 18:06:02 -07001871
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872err_restore_mac:
Jay Vosburghdd957c52007-10-09 19:57:24 -07001873 if (!bond->params.fail_over_mac) {
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001874 /* XXX TODO - fom follow mode needs to change master's
1875 * MAC if this slave's MAC is in use by the bond, or at
1876 * least print a warning.
1877 */
Moni Shoua2ab82852007-10-09 19:43:39 -07001878 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1879 addr.sa_family = slave_dev->type;
1880 dev_set_mac_address(slave_dev, &addr);
1881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00001883err_restore_mtu:
1884 dev_set_mtu(slave_dev, new_slave->original_mtu);
1885
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886err_free:
1887 kfree(new_slave);
1888
1889err_undo_flags:
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001890 bond_compute_features(bond);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001891
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 return res;
1893}
1894
1895/*
1896 * Try to release the slave device <slave> from the bond device <master>
1897 * It is legal to access curr_active_slave without a lock because all the function
1898 * is write-locked.
1899 *
1900 * The rules for slave state should be:
1901 * for Active/Backup:
1902 * Active stays on all backups go down
1903 * for Bonded connections:
1904 * The first up interface should be left on and all others downed.
1905 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08001906int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907{
Wang Chen454d7c92008-11-12 23:37:49 -08001908 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 struct slave *slave, *oldcurrent;
1910 struct sockaddr addr;
Michał Mirosławb2a103e2011-05-07 03:22:17 +00001911 u32 old_features = bond_dev->features;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912
1913 /* slave is not a slave or master is not master of this slave */
1914 if (!(slave_dev->flags & IFF_SLAVE) ||
1915 (slave_dev->master != bond_dev)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001916 pr_err("%s: Error: cannot release %s.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 bond_dev->name, slave_dev->name);
1918 return -EINVAL;
1919 }
1920
Neil Hormane843fa52010-10-13 16:01:50 +00001921 block_netpoll_tx();
Amerigo Wangdaf92092011-05-19 21:39:12 +00001922 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 write_lock_bh(&bond->lock);
1924
1925 slave = bond_get_slave_by_dev(bond, slave_dev);
1926 if (!slave) {
1927 /* not a slave of this bond */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001928 pr_info("%s: %s not enslaved\n",
1929 bond_dev->name, slave_dev->name);
Jay Vosburghf5e2a7b2006-02-07 21:17:22 -08001930 write_unlock_bh(&bond->lock);
Neil Hormane843fa52010-10-13 16:01:50 +00001931 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 return -EINVAL;
1933 }
1934
Jiri Pirko35d48902011-03-22 02:38:12 +00001935 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1936 * for this slave anymore.
1937 */
1938 netdev_rx_handler_unregister(slave_dev);
1939 write_unlock_bh(&bond->lock);
1940 synchronize_net();
1941 write_lock_bh(&bond->lock);
1942
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07001943 if (!bond->params.fail_over_mac) {
Joe Perches8e95a202009-12-03 07:58:21 +00001944 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1945 bond->slave_cnt > 1)
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001946 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1947 bond_dev->name, slave_dev->name,
1948 slave->perm_hwaddr,
1949 bond_dev->name, slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 }
1951
1952 /* Inform AD package of unbinding of slave. */
1953 if (bond->params.mode == BOND_MODE_8023AD) {
1954 /* must be called before the slave is
1955 * detached from the list
1956 */
1957 bond_3ad_unbind_slave(slave);
1958 }
1959
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001960 pr_info("%s: releasing %s interface %s\n",
1961 bond_dev->name,
Jiri Pirkoe30bc062011-03-12 03:14:37 +00001962 bond_is_active_slave(slave) ? "active" : "backup",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08001963 slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964
1965 oldcurrent = bond->curr_active_slave;
1966
1967 bond->current_arp_slave = NULL;
1968
1969 /* release the slave from its bond */
1970 bond_detach_slave(bond, slave);
1971
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001972 if (bond->primary_slave == slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 bond->primary_slave = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974
Stephen Hemminger3d632c32009-06-12 19:02:48 +00001975 if (oldcurrent == slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 bond_change_active_slave(bond, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
Holger Eitzenberger58402052008-12-09 23:07:13 -08001978 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 /* Must be called only after the slave has been
1980 * detached from the list and the curr_active_slave
1981 * has been cleared (if our_slave == old_current),
1982 * but before a new active slave is selected.
1983 */
Jay Vosburgh25433312008-01-17 16:24:59 -08001984 write_unlock_bh(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 bond_alb_deinit_slave(bond, slave);
Jay Vosburgh25433312008-01-17 16:24:59 -08001986 write_lock_bh(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 }
1988
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001989 if (oldcurrent == slave) {
1990 /*
1991 * Note that we hold RTNL over this sequence, so there
1992 * is no concern that another slave add/remove event
1993 * will interfere.
1994 */
1995 write_unlock_bh(&bond->lock);
1996 read_lock(&bond->lock);
1997 write_lock_bh(&bond->curr_slave_lock);
1998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 bond_select_active_slave(bond);
2000
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002001 write_unlock_bh(&bond->curr_slave_lock);
2002 read_unlock(&bond->lock);
2003 write_lock_bh(&bond->lock);
2004 }
2005
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 if (bond->slave_cnt == 0) {
Jay Vosburghff59c452006-03-27 13:27:43 -08002007 bond_set_carrier(bond);
2008
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 /* if the last slave was removed, zero the mac address
2010 * of the master so it will be set by the application
2011 * to the mac address of the first slave
2012 */
2013 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2014
Jiri Pirkocc0e4072011-07-20 04:54:46 +00002015 if (bond_vlan_used(bond)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002016 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2017 bond_dev->name, bond_dev->name);
2018 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2019 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 }
2022
2023 write_unlock_bh(&bond->lock);
Neil Hormane843fa52010-10-13 16:01:50 +00002024 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
Michał Mirosławb2a103e2011-05-07 03:22:17 +00002026 bond_compute_features(bond);
2027 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2028 (old_features & NETIF_F_VLAN_CHALLENGED))
2029 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2030 bond_dev->name, slave_dev->name, bond_dev->name);
2031
Mitch Williamsb76cdba2005-11-09 10:36:41 -08002032 /* must do this from outside any spinlocks */
2033 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2034
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 bond_del_vlans_from_slave(bond, slave_dev);
2036
2037 /* If the mode USES_PRIMARY, then we should only remove its
2038 * promisc and mc settings if it was the curr_active_slave, but that was
2039 * already taken care of above when we detached the slave
2040 */
2041 if (!USES_PRIMARY(bond->params.mode)) {
2042 /* unset promiscuity level from slave */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002043 if (bond_dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 dev_set_promiscuity(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
2046 /* unset allmulti level from slave */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002047 if (bond_dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 dev_set_allmulti(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
2050 /* flush master's mc_list from slave */
David S. Millerb9e40852008-07-15 00:15:08 -07002051 netif_addr_lock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 bond_mc_list_flush(bond_dev, slave_dev);
David S. Millerb9e40852008-07-15 00:15:08 -07002053 netif_addr_unlock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 }
2055
Jiri Pirko1765a572011-02-12 06:48:36 +00002056 netdev_set_bond_master(slave_dev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
Amerigo Wang8a8efa22011-02-17 23:43:32 +00002058 slave_disable_netpoll(slave);
WANG Congf6dc31a2010-05-06 00:48:51 -07002059
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 /* close slave before restoring its mac address */
2061 dev_close(slave_dev);
2062
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07002063 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
Moni Shoua2ab82852007-10-09 19:43:39 -07002064 /* restore original ("permanent") mac address */
2065 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2066 addr.sa_family = slave_dev->type;
2067 dev_set_mac_address(slave_dev, &addr);
2068 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
Jiri Pirkob15ba0f2010-05-18 05:42:40 +00002070 dev_set_mtu(slave_dev, slave->original_mtu);
2071
Jiri Pirko2d7011c2011-03-16 08:46:43 +00002072 slave_dev->priv_flags &= ~IFF_BONDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
2074 kfree(slave);
2075
2076 return 0; /* deletion OK */
2077}
2078
2079/*
Nicolas de Pesloüandadaa102011-03-19 13:36:18 -07002080* First release a slave and then destroy the bond if no more slaves are left.
Moni Shouad90a1622007-10-09 19:43:43 -07002081* Must be under rtnl_lock when this function is called.
2082*/
stephen hemminger26d8ee72010-10-15 05:09:34 +00002083static int bond_release_and_destroy(struct net_device *bond_dev,
2084 struct net_device *slave_dev)
Moni Shouad90a1622007-10-09 19:43:43 -07002085{
Wang Chen454d7c92008-11-12 23:37:49 -08002086 struct bonding *bond = netdev_priv(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07002087 int ret;
2088
2089 ret = bond_release(bond_dev, slave_dev);
2090 if ((ret == 0) && (bond->slave_cnt == 0)) {
Amerigo Wang8a8efa22011-02-17 23:43:32 +00002091 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002092 pr_info("%s: destroying bond %s.\n",
2093 bond_dev->name, bond_dev->name);
Stephen Hemminger9e716262009-06-12 19:02:47 +00002094 unregister_netdevice(bond_dev);
Moni Shouad90a1622007-10-09 19:43:43 -07002095 }
2096 return ret;
2097}
2098
2099/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 * This function releases all slaves.
2101 */
2102static int bond_release_all(struct net_device *bond_dev)
2103{
Wang Chen454d7c92008-11-12 23:37:49 -08002104 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 struct slave *slave;
2106 struct net_device *slave_dev;
2107 struct sockaddr addr;
2108
2109 write_lock_bh(&bond->lock);
2110
Jay Vosburghff59c452006-03-27 13:27:43 -08002111 netif_carrier_off(bond_dev);
2112
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002113 if (bond->slave_cnt == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115
2116 bond->current_arp_slave = NULL;
2117 bond->primary_slave = NULL;
2118 bond_change_active_slave(bond, NULL);
2119
2120 while ((slave = bond->first_slave) != NULL) {
2121 /* Inform AD package of unbinding of slave
2122 * before slave is detached from the list.
2123 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002124 if (bond->params.mode == BOND_MODE_8023AD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 bond_3ad_unbind_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
2127 slave_dev = slave->dev;
2128 bond_detach_slave(bond, slave);
2129
Jay Vosburgh25433312008-01-17 16:24:59 -08002130 /* now that the slave is detached, unlock and perform
2131 * all the undo steps that should not be called from
2132 * within a lock.
2133 */
2134 write_unlock_bh(&bond->lock);
2135
Jiri Pirko35d48902011-03-22 02:38:12 +00002136 /* unregister rx_handler early so bond_handle_frame wouldn't
2137 * be called for this slave anymore.
2138 */
2139 netdev_rx_handler_unregister(slave_dev);
2140 synchronize_net();
2141
Holger Eitzenberger58402052008-12-09 23:07:13 -08002142 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 /* must be called only after the slave
2144 * has been detached from the list
2145 */
2146 bond_alb_deinit_slave(bond, slave);
2147 }
2148
Mitch Williamsb76cdba2005-11-09 10:36:41 -08002149 bond_destroy_slave_symlinks(bond_dev, slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 bond_del_vlans_from_slave(bond, slave_dev);
2151
2152 /* If the mode USES_PRIMARY, then we should only remove its
2153 * promisc and mc settings if it was the curr_active_slave, but that was
2154 * already taken care of above when we detached the slave
2155 */
2156 if (!USES_PRIMARY(bond->params.mode)) {
2157 /* unset promiscuity level from slave */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002158 if (bond_dev->flags & IFF_PROMISC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 dev_set_promiscuity(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160
2161 /* unset allmulti level from slave */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002162 if (bond_dev->flags & IFF_ALLMULTI)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 dev_set_allmulti(slave_dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
2165 /* flush master's mc_list from slave */
David S. Millerb9e40852008-07-15 00:15:08 -07002166 netif_addr_lock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 bond_mc_list_flush(bond_dev, slave_dev);
David S. Millerb9e40852008-07-15 00:15:08 -07002168 netif_addr_unlock_bh(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 }
2170
Jiri Pirko1765a572011-02-12 06:48:36 +00002171 netdev_set_bond_master(slave_dev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172
Amerigo Wang8a8efa22011-02-17 23:43:32 +00002173 slave_disable_netpoll(slave);
2174
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 /* close slave before restoring its mac address */
2176 dev_close(slave_dev);
2177
Jay Vosburghdd957c52007-10-09 19:57:24 -07002178 if (!bond->params.fail_over_mac) {
Moni Shoua2ab82852007-10-09 19:43:39 -07002179 /* restore original ("permanent") mac address*/
2180 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2181 addr.sa_family = slave_dev->type;
2182 dev_set_mac_address(slave_dev, &addr);
2183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 kfree(slave);
2186
2187 /* re-acquire the lock before getting the next slave */
2188 write_lock_bh(&bond->lock);
2189 }
2190
2191 /* zero the mac address of the master so it will be
2192 * set by the application to the mac address of the
2193 * first slave
2194 */
2195 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2196
Jiri Pirkocc0e4072011-07-20 04:54:46 +00002197 if (bond_vlan_used(bond)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002198 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2199 bond_dev->name, bond_dev->name);
2200 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2201 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 }
2203
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002204 pr_info("%s: released all slaves\n", bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
2206out:
2207 write_unlock_bh(&bond->lock);
Michał Mirosławb2a103e2011-05-07 03:22:17 +00002208
2209 bond_compute_features(bond);
2210
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 return 0;
2212}
2213
2214/*
2215 * This function changes the active slave to slave <slave_dev>.
2216 * It returns -EINVAL in the following cases.
2217 * - <slave_dev> is not found in the list.
2218 * - There is not active slave now.
2219 * - <slave_dev> is already active.
2220 * - The link state of <slave_dev> is not BOND_LINK_UP.
2221 * - <slave_dev> is not running.
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002222 * In these cases, this function does nothing.
2223 * In the other cases, current_slave pointer is changed and 0 is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 */
2225static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2226{
Wang Chen454d7c92008-11-12 23:37:49 -08002227 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 struct slave *old_active = NULL;
2229 struct slave *new_active = NULL;
2230 int res = 0;
2231
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002232 if (!USES_PRIMARY(bond->params.mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234
2235 /* Verify that master_dev is indeed the master of slave_dev */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002236 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002239 read_lock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002241 read_lock(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 old_active = bond->curr_active_slave;
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002243 read_unlock(&bond->curr_slave_lock);
2244
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 new_active = bond_get_slave_by_dev(bond, slave_dev);
2246
2247 /*
2248 * Changing to the current active: do nothing; return success.
2249 */
2250 if (new_active && (new_active == old_active)) {
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002251 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 return 0;
2253 }
2254
2255 if ((new_active) &&
2256 (old_active) &&
2257 (new_active->link == BOND_LINK_UP) &&
2258 IS_UP(new_active->dev)) {
Neil Hormane843fa52010-10-13 16:01:50 +00002259 block_netpoll_tx();
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002260 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 bond_change_active_slave(bond, new_active);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002262 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00002263 unblock_netpoll_tx();
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002264 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 res = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002267 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268
2269 return res;
2270}
2271
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2273{
Wang Chen454d7c92008-11-12 23:37:49 -08002274 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275
2276 info->bond_mode = bond->params.mode;
2277 info->miimon = bond->params.miimon;
2278
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07002279 read_lock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 info->num_slaves = bond->slave_cnt;
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07002281 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
2283 return 0;
2284}
2285
2286static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2287{
Wang Chen454d7c92008-11-12 23:37:49 -08002288 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 struct slave *slave;
Eric Dumazet689c96c2009-04-23 03:39:04 +00002290 int i, res = -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07002292 read_lock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293
2294 bond_for_each_slave(bond, slave, i) {
2295 if (i == (int)info->slave_id) {
Eric Dumazet689c96c2009-04-23 03:39:04 +00002296 res = 0;
2297 strcpy(info->slave_name, slave->dev->name);
2298 info->link = slave->link;
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002299 info->state = bond_slave_state(slave);
Eric Dumazet689c96c2009-04-23 03:39:04 +00002300 info->link_failure_count = slave->link_failure_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 break;
2302 }
2303 }
2304
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07002305 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
Eric Dumazet689c96c2009-04-23 03:39:04 +00002307 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308}
2309
2310/*-------------------------------- Monitoring -------------------------------*/
2311
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002312
2313static int bond_miimon_inspect(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314{
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002315 struct slave *slave;
2316 int i, link_state, commit = 0;
Jiri Pirko41f89102009-04-24 03:57:29 +00002317 bool ignore_updelay;
2318
2319 ignore_updelay = !bond->curr_active_slave ? true : false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320
2321 bond_for_each_slave(bond, slave, i) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002322 slave->new_link = BOND_LINK_NOCHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002324 link_state = bond_check_dev_link(bond, slave->dev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
2326 switch (slave->link) {
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002327 case BOND_LINK_UP:
2328 if (link_state)
2329 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002331 slave->link = BOND_LINK_FAIL;
2332 slave->delay = bond->params.downdelay;
2333 if (slave->delay) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002334 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2335 bond->dev->name,
2336 (bond->params.mode ==
2337 BOND_MODE_ACTIVEBACKUP) ?
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002338 (bond_is_active_slave(slave) ?
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002339 "active " : "backup ") : "",
2340 slave->dev->name,
2341 bond->params.downdelay * bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002343 /*FALLTHRU*/
2344 case BOND_LINK_FAIL:
2345 if (link_state) {
2346 /*
2347 * recovered before downdelay expired
2348 */
2349 slave->link = BOND_LINK_UP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 slave->jiffies = jiffies;
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002351 pr_info("%s: link status up again after %d ms for interface %s.\n",
2352 bond->dev->name,
2353 (bond->params.downdelay - slave->delay) *
2354 bond->params.miimon,
2355 slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002356 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002358
2359 if (slave->delay <= 0) {
2360 slave->new_link = BOND_LINK_DOWN;
2361 commit++;
2362 continue;
2363 }
2364
2365 slave->delay--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002368 case BOND_LINK_DOWN:
2369 if (!link_state)
2370 continue;
2371
2372 slave->link = BOND_LINK_BACK;
2373 slave->delay = bond->params.updelay;
2374
2375 if (slave->delay) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002376 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2377 bond->dev->name, slave->dev->name,
2378 ignore_updelay ? 0 :
2379 bond->params.updelay *
2380 bond->params.miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002382 /*FALLTHRU*/
2383 case BOND_LINK_BACK:
2384 if (!link_state) {
2385 slave->link = BOND_LINK_DOWN;
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002386 pr_info("%s: link status down again after %d ms for interface %s.\n",
2387 bond->dev->name,
2388 (bond->params.updelay - slave->delay) *
2389 bond->params.miimon,
2390 slave->dev->name);
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002391
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002392 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 }
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002394
Jiri Pirko41f89102009-04-24 03:57:29 +00002395 if (ignore_updelay)
2396 slave->delay = 0;
2397
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002398 if (slave->delay <= 0) {
2399 slave->new_link = BOND_LINK_UP;
2400 commit++;
Jiri Pirko41f89102009-04-24 03:57:29 +00002401 ignore_updelay = false;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002402 continue;
2403 }
2404
2405 slave->delay--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 break;
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002407 }
2408 }
2409
2410 return commit;
2411}
2412
2413static void bond_miimon_commit(struct bonding *bond)
2414{
2415 struct slave *slave;
2416 int i;
2417
2418 bond_for_each_slave(bond, slave, i) {
2419 switch (slave->new_link) {
2420 case BOND_LINK_NOCHANGE:
2421 continue;
2422
2423 case BOND_LINK_UP:
2424 slave->link = BOND_LINK_UP;
2425 slave->jiffies = jiffies;
2426
2427 if (bond->params.mode == BOND_MODE_8023AD) {
2428 /* prevent it from being the active one */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002429 bond_set_backup_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002430 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2431 /* make it immediately active */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002432 bond_set_active_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002433 } else if (slave != bond->primary_slave) {
2434 /* prevent it from being the active one */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002435 bond_set_backup_slave(slave);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002436 }
2437
Krzysztof Piotr Oledzki546add72010-10-06 14:28:22 -07002438 bond_update_speed_duplex(slave);
2439
David Decotigny5d305302011-04-13 15:22:31 +00002440 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
Krzysztof Piotr Oledzki700c2a72010-10-06 14:25:06 -07002441 bond->dev->name, slave->dev->name,
2442 slave->speed, slave->duplex ? "full" : "half");
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002443
2444 /* notify ad that the link status has changed */
2445 if (bond->params.mode == BOND_MODE_8023AD)
2446 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2447
Holger Eitzenberger58402052008-12-09 23:07:13 -08002448 if (bond_is_lb(bond))
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002449 bond_alb_handle_link_change(bond, slave,
2450 BOND_LINK_UP);
2451
2452 if (!bond->curr_active_slave ||
2453 (slave == bond->primary_slave))
2454 goto do_failover;
2455
2456 continue;
2457
2458 case BOND_LINK_DOWN:
Jay Vosburghfba4acd2008-10-30 17:41:14 -07002459 if (slave->link_failure_count < UINT_MAX)
2460 slave->link_failure_count++;
2461
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002462 slave->link = BOND_LINK_DOWN;
2463
2464 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2465 bond->params.mode == BOND_MODE_8023AD)
2466 bond_set_slave_inactive_flags(slave);
2467
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002468 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2469 bond->dev->name, slave->dev->name);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002470
2471 if (bond->params.mode == BOND_MODE_8023AD)
2472 bond_3ad_handle_link_change(slave,
2473 BOND_LINK_DOWN);
2474
Jiri Pirkoae63e802009-05-27 05:42:36 +00002475 if (bond_is_lb(bond))
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002476 bond_alb_handle_link_change(bond, slave,
2477 BOND_LINK_DOWN);
2478
2479 if (slave == bond->curr_active_slave)
2480 goto do_failover;
2481
2482 continue;
2483
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 default:
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002485 pr_err("%s: invalid new link %d on slave %s\n",
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002486 bond->dev->name, slave->new_link,
2487 slave->dev->name);
2488 slave->new_link = BOND_LINK_NOCHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002490 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 }
2492
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002493do_failover:
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002494 ASSERT_RTNL();
Neil Hormane843fa52010-10-13 16:01:50 +00002495 block_netpoll_tx();
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002496 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 bond_select_active_slave(bond);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002498 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00002499 unblock_netpoll_tx();
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002500 }
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002501
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002502 bond_set_carrier(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503}
2504
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002505/*
2506 * bond_mii_monitor
2507 *
2508 * Really a wrapper that splits the mii monitor into two phases: an
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002509 * inspection, then (if inspection indicates something needs to be done)
2510 * an acquisition of appropriate locks followed by a commit phase to
2511 * implement whatever link state changes are indicated.
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002512 */
2513void bond_mii_monitor(struct work_struct *work)
2514{
2515 struct bonding *bond = container_of(work, struct bonding,
2516 mii_work.work);
Ben Hutchingsad246c92011-04-26 15:25:52 +00002517 bool should_notify_peers = false;
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002518
2519 read_lock(&bond->lock);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002520 if (bond->kill_timers)
2521 goto out;
2522
2523 if (bond->slave_cnt == 0)
2524 goto re_arm;
Jay Vosburghb59f9f72008-06-13 18:12:03 -07002525
Ben Hutchingsad246c92011-04-26 15:25:52 +00002526 should_notify_peers = bond_should_notify_peers(bond);
2527
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002528 if (bond_miimon_inspect(bond)) {
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002529 read_unlock(&bond->lock);
2530 rtnl_lock();
2531 read_lock(&bond->lock);
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002532
2533 bond_miimon_commit(bond);
2534
Jay Vosburgh56556622008-01-17 16:25:03 -08002535 read_unlock(&bond->lock);
2536 rtnl_unlock(); /* might sleep, hold no other locks */
2537 read_lock(&bond->lock);
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002538 }
2539
Jay Vosburghf0c76d62008-07-02 18:21:58 -07002540re_arm:
2541 if (bond->params.miimon)
2542 queue_delayed_work(bond->wq, &bond->mii_work,
2543 msecs_to_jiffies(bond->params.miimon));
2544out:
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002545 read_unlock(&bond->lock);
Ben Hutchingsad246c92011-04-26 15:25:52 +00002546
2547 if (should_notify_peers) {
2548 rtnl_lock();
2549 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2550 rtnl_unlock();
2551 }
Jay Vosburgh0b0eef62007-10-17 17:37:48 -07002552}
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002553
Al Virod3bb52b2007-08-22 20:06:58 -04002554static __be32 bond_glean_dev_ip(struct net_device *dev)
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002555{
2556 struct in_device *idev;
2557 struct in_ifaddr *ifa;
Al Viroa144ea42006-09-28 18:00:55 -07002558 __be32 addr = 0;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002559
2560 if (!dev)
2561 return 0;
2562
2563 rcu_read_lock();
Herbert Xue5ed6392005-10-03 14:35:55 -07002564 idev = __in_dev_get_rcu(dev);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002565 if (!idev)
2566 goto out;
2567
2568 ifa = idev->ifa_list;
2569 if (!ifa)
2570 goto out;
2571
2572 addr = ifa->ifa_local;
2573out:
2574 rcu_read_unlock();
2575 return addr;
2576}
2577
Al Virod3bb52b2007-08-22 20:06:58 -04002578static int bond_has_this_ip(struct bonding *bond, __be32 ip)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002579{
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07002580 struct vlan_entry *vlan;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002581
2582 if (ip == bond->master_ip)
2583 return 1;
2584
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07002585 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002586 if (ip == vlan->vlan_ip)
2587 return 1;
2588 }
2589
2590 return 0;
2591}
2592
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002593/*
2594 * We go to the (large) trouble of VLAN tagging ARP frames because
2595 * switches in VLAN mode (especially if ports are configured as
2596 * "native" to a VLAN) might not pass non-tagged frames.
2597 */
Al Virod3bb52b2007-08-22 20:06:58 -04002598static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002599{
2600 struct sk_buff *skb;
2601
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002602 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002603 slave_dev->name, dest_ip, src_ip, vlan_id);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002604
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002605 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2606 NULL, slave_dev->dev_addr, NULL);
2607
2608 if (!skb) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002609 pr_err("ARP packet allocation failed\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002610 return;
2611 }
2612 if (vlan_id) {
2613 skb = vlan_put_tag(skb, vlan_id);
2614 if (!skb) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002615 pr_err("failed to insert VLAN tag\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002616 return;
2617 }
2618 }
2619 arp_xmit(skb);
2620}
2621
2622
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2624{
David S. Millerb23dd4f2011-03-02 14:31:35 -08002625 int i, vlan_id;
Al Virod3bb52b2007-08-22 20:06:58 -04002626 __be32 *targets = bond->params.arp_targets;
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07002627 struct vlan_entry *vlan;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002628 struct net_device *vlan_dev;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002629 struct rtable *rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630
Mitch Williams6b780562005-11-09 10:36:19 -08002631 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2632 if (!targets[i])
Brian Haley5a31bec2009-04-13 00:11:30 -07002633 break;
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002634 pr_debug("basa: target %x\n", targets[i]);
Jiri Pirkocc0e4072011-07-20 04:54:46 +00002635 if (!bond_vlan_used(bond)) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002636 pr_debug("basa: empty vlan: arp_send\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002637 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2638 bond->master_ip, 0);
2639 continue;
2640 }
2641
2642 /*
2643 * If VLANs are configured, we do a route lookup to
2644 * determine which VLAN interface would be used, so we
2645 * can tag the ARP with the proper VLAN tag.
2646 */
David S. Miller78fbfd82011-03-12 00:00:52 -05002647 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2648 RTO_ONLINK, 0);
David S. Millerb23dd4f2011-03-02 14:31:35 -08002649 if (IS_ERR(rt)) {
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002650 if (net_ratelimit()) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002651 pr_warning("%s: no route to arp_ip_target %pI4\n",
David S. Miller78fbfd82011-03-12 00:00:52 -05002652 bond->dev->name, &targets[i]);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002653 }
2654 continue;
2655 }
2656
2657 /*
2658 * This target is not on a VLAN
2659 */
Changli Gaod8d1f302010-06-10 23:31:35 -07002660 if (rt->dst.dev == bond->dev) {
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002661 ip_rt_put(rt);
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002662 pr_debug("basa: rtdev == bond->dev: arp_send\n");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002663 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2664 bond->master_ip, 0);
2665 continue;
2666 }
2667
2668 vlan_id = 0;
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07002669 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Jiri Pirkocc0e4072011-07-20 04:54:46 +00002670 rcu_read_lock();
2671 vlan_dev = __vlan_find_dev_deep(bond->dev,
2672 vlan->vlan_id);
2673 rcu_read_unlock();
Changli Gaod8d1f302010-06-10 23:31:35 -07002674 if (vlan_dev == rt->dst.dev) {
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002675 vlan_id = vlan->vlan_id;
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002676 pr_debug("basa: vlan match on %s %d\n",
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002677 vlan_dev->name, vlan_id);
2678 break;
2679 }
2680 }
2681
2682 if (vlan_id) {
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002683 ip_rt_put(rt);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002684 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2685 vlan->vlan_ip, vlan_id);
2686 continue;
2687 }
2688
2689 if (net_ratelimit()) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002690 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
David S. Miller78fbfd82011-03-12 00:00:52 -05002691 bond->dev->name, &targets[i],
Changli Gaod8d1f302010-06-10 23:31:35 -07002692 rt->dst.dev ? rt->dst.dev->name : "NULL");
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002693 }
Jay Vosburghed4b9f82005-09-14 14:52:09 -07002694 ip_rt_put(rt);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002695 }
2696}
2697
Al Virod3bb52b2007-08-22 20:06:58 -04002698static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002699{
2700 int i;
Al Virod3bb52b2007-08-22 20:06:58 -04002701 __be32 *targets = bond->params.arp_targets;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002702
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002703 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002704 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002705 &sip, &tip, i, &targets[i],
2706 bond_has_this_ip(bond, tip));
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002707 if (sip == targets[i]) {
2708 if (bond_has_this_ip(bond, tip))
2709 slave->last_arp_rx = jiffies;
2710 return;
2711 }
2712 }
2713}
2714
Jiri Pirko3aba8912011-04-19 03:48:16 +00002715static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2716 struct slave *slave)
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002717{
2718 struct arphdr *arp;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002719 unsigned char *arp_ptr;
Al Virod3bb52b2007-08-22 20:06:58 -04002720 __be32 sip, tip;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002721
Jiri Pirko3aba8912011-04-19 03:48:16 +00002722 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2723 return;
Andy Gospodarek1f3c8802009-12-14 10:48:58 +00002724
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002725 read_lock(&bond->lock);
2726
Jiri Pirko3aba8912011-04-19 03:48:16 +00002727 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2728 bond->dev->name, skb->dev->name);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002729
Jiri Pirko3aba8912011-04-19 03:48:16 +00002730 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002731 goto out_unlock;
2732
Arnaldo Carvalho de Melod0a92be2007-03-12 20:56:31 -03002733 arp = arp_hdr(skb);
Jiri Pirko3aba8912011-04-19 03:48:16 +00002734 if (arp->ar_hln != bond->dev->addr_len ||
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002735 skb->pkt_type == PACKET_OTHERHOST ||
2736 skb->pkt_type == PACKET_LOOPBACK ||
2737 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2738 arp->ar_pro != htons(ETH_P_IP) ||
2739 arp->ar_pln != 4)
2740 goto out_unlock;
2741
2742 arp_ptr = (unsigned char *)(arp + 1);
Jiri Pirko3aba8912011-04-19 03:48:16 +00002743 arp_ptr += bond->dev->addr_len;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002744 memcpy(&sip, arp_ptr, 4);
Jiri Pirko3aba8912011-04-19 03:48:16 +00002745 arp_ptr += 4 + bond->dev->addr_len;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002746 memcpy(&tip, arp_ptr, 4);
2747
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08002748 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002749 bond->dev->name, slave->dev->name, bond_slave_state(slave),
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002750 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2751 &sip, &tip);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002752
2753 /*
2754 * Backup slaves won't see the ARP reply, but do come through
2755 * here for each ARP probe (so we swap the sip/tip to validate
2756 * the probe). In a "redundant switch, common router" type of
2757 * configuration, the ARP probe will (hopefully) travel from
2758 * the active, through one switch, the router, then the other
2759 * switch before reaching the backup.
2760 */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002761 if (bond_is_active_slave(slave))
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002762 bond_validate_arp(bond, slave, sip, tip);
2763 else
2764 bond_validate_arp(bond, slave, tip, sip);
2765
2766out_unlock:
2767 read_unlock(&bond->lock);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07002768}
2769
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770/*
2771 * this function is called regularly to monitor each slave's link
2772 * ensuring that traffic is being sent and received when arp monitoring
2773 * is used in load-balancing mode. if the adapter has been dormant, then an
2774 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2775 * arp monitoring in active backup mode.
2776 */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07002777void bond_loadbalance_arp_mon(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778{
Jay Vosburgh1b76b312007-10-17 17:37:45 -07002779 struct bonding *bond = container_of(work, struct bonding,
2780 arp_work.work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 struct slave *slave, *oldcurrent;
2782 int do_failover = 0;
2783 int delta_in_ticks;
2784 int i;
2785
2786 read_lock(&bond->lock);
2787
Jay Vosburgh5ce0da82008-05-17 21:10:07 -07002788 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002790 if (bond->kill_timers)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002793 if (bond->slave_cnt == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 goto re_arm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795
2796 read_lock(&bond->curr_slave_lock);
2797 oldcurrent = bond->curr_active_slave;
2798 read_unlock(&bond->curr_slave_lock);
2799
2800 /* see if any of the previous devices are up now (i.e. they have
2801 * xmt and rcv traffic). the curr_active_slave does not come into
2802 * the picture unless it is null. also, slave->jiffies is not needed
2803 * here because we send an arp on each slave and give a slave as
2804 * long as it needs to get the tx/rx within the delta.
2805 * TODO: what about up/down delay in arp mode? it wasn't here before
2806 * so it can wait
2807 */
2808 bond_for_each_slave(bond, slave, i) {
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002809 unsigned long trans_start = dev_trans_start(slave->dev);
2810
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 if (slave->link != BOND_LINK_UP) {
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002812 if (time_in_range(jiffies,
2813 trans_start - delta_in_ticks,
2814 trans_start + delta_in_ticks) &&
2815 time_in_range(jiffies,
2816 slave->dev->last_rx - delta_in_ticks,
2817 slave->dev->last_rx + delta_in_ticks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818
2819 slave->link = BOND_LINK_UP;
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002820 bond_set_active_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821
2822 /* primary_slave has no meaning in round-robin
2823 * mode. the window of a slave being up and
2824 * curr_active_slave being null after enslaving
2825 * is closed.
2826 */
2827 if (!oldcurrent) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002828 pr_info("%s: link status definitely up for interface %s, ",
2829 bond->dev->name,
2830 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 do_failover = 1;
2832 } else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002833 pr_info("%s: interface %s is now up\n",
2834 bond->dev->name,
2835 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 }
2837 }
2838 } else {
2839 /* slave->link == BOND_LINK_UP */
2840
2841 /* not all switches will respond to an arp request
2842 * when the source ip is 0, so don't take the link down
2843 * if we don't know our ip yet
2844 */
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002845 if (!time_in_range(jiffies,
2846 trans_start - delta_in_ticks,
2847 trans_start + 2 * delta_in_ticks) ||
2848 !time_in_range(jiffies,
2849 slave->dev->last_rx - delta_in_ticks,
2850 slave->dev->last_rx + 2 * delta_in_ticks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851
2852 slave->link = BOND_LINK_DOWN;
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002853 bond_set_backup_slave(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002855 if (slave->link_failure_count < UINT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857
Joe Perchesa4aee5c2009-12-13 20:06:07 -08002858 pr_info("%s: interface %s is now down.\n",
2859 bond->dev->name,
2860 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002862 if (slave == oldcurrent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 do_failover = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 }
2865 }
2866
2867 /* note: if switch is in round-robin mode, all links
2868 * must tx arp to ensure all links rx an arp - otherwise
2869 * links may oscillate or not come up at all; if switch is
2870 * in something like xor mode, there is nothing we can
2871 * do - all replies will be rx'ed on same link causing slaves
2872 * to be unstable during low/no traffic periods
2873 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00002874 if (IS_UP(slave->dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 bond_arp_send_all(bond, slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 }
2877
2878 if (do_failover) {
Neil Hormane843fa52010-10-13 16:01:50 +00002879 block_netpoll_tx();
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002880 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881
2882 bond_select_active_slave(bond);
2883
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07002884 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00002885 unblock_netpoll_tx();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 }
2887
2888re_arm:
Jay Vosburgh1b76b312007-10-17 17:37:45 -07002889 if (bond->params.arp_interval)
2890 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891out:
2892 read_unlock(&bond->lock);
2893}
2894
2895/*
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002896 * Called to inspect slaves for active-backup mode ARP monitor link state
2897 * changes. Sets new_link in slaves to specify what action should take
2898 * place for the slave. Returns 0 if no changes are found, >0 if changes
2899 * to link states must be committed.
2900 *
2901 * Called with bond->lock held for read.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 */
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002903static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 struct slave *slave;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002906 int i, commit = 0;
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002907 unsigned long trans_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 bond_for_each_slave(bond, slave, i) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002910 slave->new_link = BOND_LINK_NOCHANGE;
2911
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 if (slave->link != BOND_LINK_UP) {
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002913 if (time_in_range(jiffies,
2914 slave_last_rx(bond, slave) - delta_in_ticks,
2915 slave_last_rx(bond, slave) + delta_in_ticks)) {
2916
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002917 slave->new_link = BOND_LINK_UP;
2918 commit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002921 continue;
2922 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002924 /*
2925 * Give slaves 2*delta after being enslaved or made
2926 * active. This avoids bouncing, as the last receive
2927 * times need a full ARP monitor cycle to be updated.
2928 */
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002929 if (time_in_range(jiffies,
2930 slave->jiffies - delta_in_ticks,
2931 slave->jiffies + 2 * delta_in_ticks))
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002932 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002934 /*
2935 * Backup slave is down if:
2936 * - No current_arp_slave AND
2937 * - more than 3*delta since last receive AND
2938 * - the bond has an IP address
2939 *
2940 * Note: a non-null current_arp_slave indicates
2941 * the curr_active_slave went down and we are
2942 * searching for a new one; under this condition
2943 * we only take the curr_active_slave down - this
2944 * gives each slave a chance to tx/rx traffic
2945 * before being taken out
2946 */
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002947 if (!bond_is_active_slave(slave) &&
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002948 !bond->current_arp_slave &&
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002949 !time_in_range(jiffies,
2950 slave_last_rx(bond, slave) - delta_in_ticks,
2951 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2952
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002953 slave->new_link = BOND_LINK_DOWN;
2954 commit++;
2955 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002957 /*
2958 * Active slave is down if:
2959 * - more than 2*delta since transmitting OR
2960 * - (more than 2*delta since receive AND
2961 * the bond has an IP address)
2962 */
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002963 trans_start = dev_trans_start(slave->dev);
Jiri Pirkoe30bc062011-03-12 03:14:37 +00002964 if (bond_is_active_slave(slave) &&
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002965 (!time_in_range(jiffies,
2966 trans_start - delta_in_ticks,
2967 trans_start + 2 * delta_in_ticks) ||
2968 !time_in_range(jiffies,
2969 slave_last_rx(bond, slave) - delta_in_ticks,
2970 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2971
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002972 slave->new_link = BOND_LINK_DOWN;
2973 commit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 }
2975 }
2976
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002977 return commit;
2978}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002980/*
2981 * Called to commit link state changes noted by inspection step of
2982 * active-backup mode ARP monitor.
2983 *
2984 * Called with RTNL and bond->lock for read.
2985 */
2986static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2987{
2988 struct slave *slave;
2989 int i;
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002990 unsigned long trans_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991
Jay Vosburghb2220ca2008-05-17 21:10:13 -07002992 bond_for_each_slave(bond, slave, i) {
2993 switch (slave->new_link) {
2994 case BOND_LINK_NOCHANGE:
2995 continue;
2996
2997 case BOND_LINK_UP:
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00002998 trans_start = dev_trans_start(slave->dev);
Jiri Pirkob9f60252009-08-31 11:09:38 +00002999 if ((!bond->curr_active_slave &&
Jiri Bohaccb32f2a2010-09-02 05:45:54 +00003000 time_in_range(jiffies,
3001 trans_start - delta_in_ticks,
3002 trans_start + delta_in_ticks)) ||
Jiri Pirkob9f60252009-08-31 11:09:38 +00003003 bond->curr_active_slave != slave) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003004 slave->link = BOND_LINK_UP;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003005 bond->current_arp_slave = NULL;
3006
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003007 pr_info("%s: link status definitely up for interface %s.\n",
Jiri Pirkob9f60252009-08-31 11:09:38 +00003008 bond->dev->name, slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003009
Jiri Pirkob9f60252009-08-31 11:09:38 +00003010 if (!bond->curr_active_slave ||
3011 (slave == bond->primary_slave))
3012 goto do_failover;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003013
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003014 }
3015
Jiri Pirkob9f60252009-08-31 11:09:38 +00003016 continue;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003017
3018 case BOND_LINK_DOWN:
3019 if (slave->link_failure_count < UINT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003022 slave->link = BOND_LINK_DOWN;
Jiri Pirkob9f60252009-08-31 11:09:38 +00003023 bond_set_slave_inactive_flags(slave);
3024
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003025 pr_info("%s: link status definitely down for interface %s, disabling it\n",
Jiri Pirkob9f60252009-08-31 11:09:38 +00003026 bond->dev->name, slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003027
3028 if (slave == bond->curr_active_slave) {
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003029 bond->current_arp_slave = NULL;
Jiri Pirkob9f60252009-08-31 11:09:38 +00003030 goto do_failover;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003031 }
Jiri Pirkob9f60252009-08-31 11:09:38 +00003032
3033 continue;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003034
3035 default:
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003036 pr_err("%s: impossible: new_link %d on slave %s\n",
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003037 bond->dev->name, slave->new_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 slave->dev->name);
Jiri Pirkob9f60252009-08-31 11:09:38 +00003039 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041
Jiri Pirkob9f60252009-08-31 11:09:38 +00003042do_failover:
3043 ASSERT_RTNL();
Neil Hormane843fa52010-10-13 16:01:50 +00003044 block_netpoll_tx();
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003045 write_lock_bh(&bond->curr_slave_lock);
Jiri Pirkob9f60252009-08-31 11:09:38 +00003046 bond_select_active_slave(bond);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003047 write_unlock_bh(&bond->curr_slave_lock);
Neil Hormane843fa52010-10-13 16:01:50 +00003048 unblock_netpoll_tx();
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003049 }
3050
3051 bond_set_carrier(bond);
3052}
3053
3054/*
3055 * Send ARP probes for active-backup mode ARP monitor.
3056 *
3057 * Called with bond->lock held for read.
3058 */
3059static void bond_ab_arp_probe(struct bonding *bond)
3060{
3061 struct slave *slave;
3062 int i;
3063
3064 read_lock(&bond->curr_slave_lock);
3065
3066 if (bond->current_arp_slave && bond->curr_active_slave)
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003067 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3068 bond->current_arp_slave->dev->name,
3069 bond->curr_active_slave->dev->name);
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003070
3071 if (bond->curr_active_slave) {
3072 bond_arp_send_all(bond, bond->curr_active_slave);
3073 read_unlock(&bond->curr_slave_lock);
3074 return;
3075 }
3076
3077 read_unlock(&bond->curr_slave_lock);
3078
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 /* if we don't have a curr_active_slave, search for the next available
3080 * backup slave from the current_arp_slave and make it the candidate
3081 * for becoming the curr_active_slave
3082 */
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003083
3084 if (!bond->current_arp_slave) {
3085 bond->current_arp_slave = bond->first_slave;
3086 if (!bond->current_arp_slave)
3087 return;
3088 }
3089
3090 bond_set_slave_inactive_flags(bond->current_arp_slave);
3091
3092 /* search for next candidate */
3093 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3094 if (IS_UP(slave->dev)) {
3095 slave->link = BOND_LINK_BACK;
3096 bond_set_slave_active_flags(slave);
3097 bond_arp_send_all(bond, slave);
3098 slave->jiffies = jiffies;
3099 bond->current_arp_slave = slave;
3100 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 }
3102
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003103 /* if the link state is up at this point, we
3104 * mark it down - this can happen if we have
3105 * simultaneous link failures and
3106 * reselect_active_interface doesn't make this
3107 * one the current slave so it is still marked
3108 * up when it is actually down
3109 */
3110 if (slave->link == BOND_LINK_UP) {
3111 slave->link = BOND_LINK_DOWN;
3112 if (slave->link_failure_count < UINT_MAX)
3113 slave->link_failure_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003115 bond_set_slave_inactive_flags(slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003117 pr_info("%s: backup interface %s is now down.\n",
3118 bond->dev->name, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119 }
3120 }
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003121}
3122
3123void bond_activebackup_arp_mon(struct work_struct *work)
3124{
3125 struct bonding *bond = container_of(work, struct bonding,
3126 arp_work.work);
Ben Hutchingsad246c92011-04-26 15:25:52 +00003127 bool should_notify_peers = false;
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003128 int delta_in_ticks;
3129
3130 read_lock(&bond->lock);
3131
3132 if (bond->kill_timers)
3133 goto out;
3134
3135 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3136
3137 if (bond->slave_cnt == 0)
3138 goto re_arm;
3139
Ben Hutchingsad246c92011-04-26 15:25:52 +00003140 should_notify_peers = bond_should_notify_peers(bond);
3141
Jay Vosburghb2220ca2008-05-17 21:10:13 -07003142 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3143 read_unlock(&bond->lock);
3144 rtnl_lock();
3145 read_lock(&bond->lock);
3146
3147 bond_ab_arp_commit(bond, delta_in_ticks);
3148
3149 read_unlock(&bond->lock);
3150 rtnl_unlock();
3151 read_lock(&bond->lock);
3152 }
3153
3154 bond_ab_arp_probe(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155
3156re_arm:
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003157 if (bond->params.arp_interval)
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003158 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159out:
3160 read_unlock(&bond->lock);
Ben Hutchingsad246c92011-04-26 15:25:52 +00003161
3162 if (should_notify_peers) {
3163 rtnl_lock();
3164 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3165 rtnl_unlock();
3166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167}
3168
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169/*-------------------------- netdev event handling --------------------------*/
3170
3171/*
3172 * Change device name
3173 */
3174static int bond_event_changename(struct bonding *bond)
3175{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176 bond_remove_proc_entry(bond);
3177 bond_create_proc_entry(bond);
Stephen Hemminger7e083842009-06-12 19:02:46 +00003178
Taku Izumif073c7c2010-12-09 15:17:13 +00003179 bond_debug_reregister(bond);
3180
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 return NOTIFY_DONE;
3182}
3183
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003184static int bond_master_netdev_event(unsigned long event,
3185 struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186{
Wang Chen454d7c92008-11-12 23:37:49 -08003187 struct bonding *event_bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188
3189 switch (event) {
3190 case NETDEV_CHANGENAME:
3191 return bond_event_changename(event_bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 default:
3193 break;
3194 }
3195
3196 return NOTIFY_DONE;
3197}
3198
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003199static int bond_slave_netdev_event(unsigned long event,
3200 struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201{
3202 struct net_device *bond_dev = slave_dev->master;
Wang Chen454d7c92008-11-12 23:37:49 -08003203 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204
3205 switch (event) {
3206 case NETDEV_UNREGISTER:
3207 if (bond_dev) {
Jay Vosburgh1284cd32007-10-15 16:44:27 -07003208 if (bond->setup_by_slave)
3209 bond_release_and_destroy(bond_dev, slave_dev);
3210 else
3211 bond_release(bond_dev, slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212 }
3213 break;
3214 case NETDEV_CHANGE:
Jay Vosburgh17d04502009-03-18 18:38:25 -07003215 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3216 struct slave *slave;
3217
3218 slave = bond_get_slave_by_dev(bond, slave_dev);
3219 if (slave) {
David Decotigny5d305302011-04-13 15:22:31 +00003220 u32 old_speed = slave->speed;
David Decotigny65cce192011-04-13 15:22:30 +00003221 u8 old_duplex = slave->duplex;
Jay Vosburgh17d04502009-03-18 18:38:25 -07003222
3223 bond_update_speed_duplex(slave);
3224
3225 if (bond_is_lb(bond))
3226 break;
3227
3228 if (old_speed != slave->speed)
3229 bond_3ad_adapter_speed_changed(slave);
3230 if (old_duplex != slave->duplex)
3231 bond_3ad_adapter_duplex_changed(slave);
3232 }
3233 }
3234
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235 break;
3236 case NETDEV_DOWN:
3237 /*
3238 * ... Or is it this?
3239 */
3240 break;
3241 case NETDEV_CHANGEMTU:
3242 /*
3243 * TODO: Should slaves be allowed to
3244 * independently alter their MTU? For
3245 * an active-backup bond, slaves need
3246 * not be the same type of device, so
3247 * MTUs may vary. For other modes,
3248 * slaves arguably should have the
3249 * same MTUs. To do this, we'd need to
3250 * take over the slave's change_mtu
3251 * function for the duration of their
3252 * servitude.
3253 */
3254 break;
3255 case NETDEV_CHANGENAME:
3256 /*
3257 * TODO: handle changing the primary's name
3258 */
3259 break;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04003260 case NETDEV_FEAT_CHANGE:
3261 bond_compute_features(bond);
3262 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 default:
3264 break;
3265 }
3266
3267 return NOTIFY_DONE;
3268}
3269
3270/*
3271 * bond_netdev_event: handle netdev notifier chain events.
3272 *
3273 * This function receives events for the netdev chain. The caller (an
Alan Sterne041c682006-03-27 01:16:30 -08003274 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 * locks for us to safely manipulate the slave devices (RTNL lock,
3276 * dev_probe_lock).
3277 */
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003278static int bond_netdev_event(struct notifier_block *this,
3279 unsigned long event, void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280{
3281 struct net_device *event_dev = (struct net_device *)ptr;
3282
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003283 pr_debug("event_dev: %s, event: %lx\n",
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003284 event_dev ? event_dev->name : "None",
3285 event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286
Jay Vosburgh0b680e72006-09-22 21:54:10 -07003287 if (!(event_dev->priv_flags & IFF_BONDING))
3288 return NOTIFY_DONE;
3289
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290 if (event_dev->flags & IFF_MASTER) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003291 pr_debug("IFF_MASTER\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292 return bond_master_netdev_event(event, event_dev);
3293 }
3294
3295 if (event_dev->flags & IFF_SLAVE) {
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003296 pr_debug("IFF_SLAVE\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 return bond_slave_netdev_event(event, event_dev);
3298 }
3299
3300 return NOTIFY_DONE;
3301}
3302
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003303/*
3304 * bond_inetaddr_event: handle inetaddr notifier chain events.
3305 *
3306 * We keep track of device IPs primarily to use as source addresses in
3307 * ARP monitor probes (rather than spewing out broadcasts all the time).
3308 *
3309 * We track one IP for the main device (if it has one), plus one per VLAN.
3310 */
3311static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3312{
3313 struct in_ifaddr *ifa = ptr;
3314 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00003315 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07003316 struct bonding *bond;
3317 struct vlan_entry *vlan;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003318
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00003319 list_for_each_entry(bond, &bn->dev_list, bond_list) {
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003320 if (bond->dev == event_dev) {
3321 switch (event) {
3322 case NETDEV_UP:
3323 bond->master_ip = ifa->ifa_local;
3324 return NOTIFY_OK;
3325 case NETDEV_DOWN:
3326 bond->master_ip = bond_glean_dev_ip(bond->dev);
3327 return NOTIFY_OK;
3328 default:
3329 return NOTIFY_DONE;
3330 }
3331 }
3332
Pavel Emelyanov0883bec2008-05-17 21:10:10 -07003333 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
Jiri Pirkocc0e4072011-07-20 04:54:46 +00003334 vlan_dev = __vlan_find_dev_deep(bond->dev,
3335 vlan->vlan_id);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003336 if (vlan_dev == event_dev) {
3337 switch (event) {
3338 case NETDEV_UP:
3339 vlan->vlan_ip = ifa->ifa_local;
3340 return NOTIFY_OK;
3341 case NETDEV_DOWN:
3342 vlan->vlan_ip =
3343 bond_glean_dev_ip(vlan_dev);
3344 return NOTIFY_OK;
3345 default:
3346 return NOTIFY_DONE;
3347 }
3348 }
3349 }
3350 }
3351 return NOTIFY_DONE;
3352}
3353
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354static struct notifier_block bond_netdev_notifier = {
3355 .notifier_call = bond_netdev_event,
3356};
3357
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003358static struct notifier_block bond_inetaddr_notifier = {
3359 .notifier_call = bond_inetaddr_event,
3360};
3361
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003362/*---------------------------- Hashing Policies -----------------------------*/
3363
3364/*
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08003365 * Hash for the output device based upon layer 2 and layer 3 data. If
3366 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3367 */
Jasper Spaansa361c832009-10-23 04:09:24 +00003368static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08003369{
3370 struct ethhdr *data = (struct ethhdr *)skb->data;
3371 struct iphdr *iph = ip_hdr(skb);
3372
Brian Haleyf14c4e42008-09-02 10:08:08 -04003373 if (skb->protocol == htons(ETH_P_IP)) {
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08003374 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
Jasper Spaansd3da6832009-10-23 04:08:46 +00003375 (data->h_dest[5] ^ data->h_source[5])) % count;
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08003376 }
3377
Jasper Spaansd3da6832009-10-23 04:08:46 +00003378 return (data->h_dest[5] ^ data->h_source[5]) % count;
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08003379}
3380
3381/*
Michael Opdenacker59c51592007-05-09 08:57:56 +02003382 * Hash for the output device based upon layer 3 and layer 4 data. If
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003383 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3384 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3385 */
Jasper Spaansa361c832009-10-23 04:09:24 +00003386static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003387{
3388 struct ethhdr *data = (struct ethhdr *)skb->data;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07003389 struct iphdr *iph = ip_hdr(skb);
Al Virod3bb52b2007-08-22 20:06:58 -04003390 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003391 int layer4_xor = 0;
3392
Brian Haleyf14c4e42008-09-02 10:08:08 -04003393 if (skb->protocol == htons(ETH_P_IP)) {
Paul Gortmaker56f8a752011-06-21 20:33:34 -07003394 if (!ip_is_fragment(iph) &&
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003395 (iph->protocol == IPPROTO_TCP ||
3396 iph->protocol == IPPROTO_UDP)) {
Al Virod3bb52b2007-08-22 20:06:58 -04003397 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003398 }
3399 return (layer4_xor ^
3400 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3401
3402 }
3403
Jasper Spaansd3da6832009-10-23 04:08:46 +00003404 return (data->h_dest[5] ^ data->h_source[5]) % count;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003405}
3406
3407/*
3408 * Hash for the output device based upon layer 2 data
3409 */
Jasper Spaansa361c832009-10-23 04:09:24 +00003410static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003411{
3412 struct ethhdr *data = (struct ethhdr *)skb->data;
3413
Jasper Spaansd3da6832009-10-23 04:08:46 +00003414 return (data->h_dest[5] ^ data->h_source[5]) % count;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04003415}
3416
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417/*-------------------------- Device entry points ----------------------------*/
3418
3419static int bond_open(struct net_device *bond_dev)
3420{
Wang Chen454d7c92008-11-12 23:37:49 -08003421 struct bonding *bond = netdev_priv(bond_dev);
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003422 struct slave *slave;
3423 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424
3425 bond->kill_timers = 0;
3426
Peter Pan(潘卫平)ba3211c2011-08-15 15:57:35 +00003427 /* reset slave->backup and slave->inactive */
3428 read_lock(&bond->lock);
3429 if (bond->slave_cnt > 0) {
3430 read_lock(&bond->curr_slave_lock);
3431 bond_for_each_slave(bond, slave, i) {
3432 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3433 && (slave != bond->curr_active_slave)) {
3434 bond_set_slave_inactive_flags(slave);
3435 } else {
3436 bond_set_slave_active_flags(slave);
3437 }
3438 }
3439 read_unlock(&bond->curr_slave_lock);
3440 }
3441 read_unlock(&bond->lock);
3442
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00003443 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3444
Holger Eitzenberger58402052008-12-09 23:07:13 -08003445 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446 /* bond_alb_initialize must be called before the timer
3447 * is started.
3448 */
3449 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3450 /* something went wrong - fail the open operation */
stephen hemmingerb4739462010-01-25 23:34:15 +00003451 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452 }
3453
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003454 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3455 queue_delayed_work(bond->wq, &bond->alb_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003456 }
3457
3458 if (bond->params.miimon) { /* link check interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003459 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3460 queue_delayed_work(bond->wq, &bond->mii_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461 }
3462
3463 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003464 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3465 INIT_DELAYED_WORK(&bond->arp_work,
3466 bond_activebackup_arp_mon);
3467 else
3468 INIT_DELAYED_WORK(&bond->arp_work,
3469 bond_loadbalance_arp_mon);
3470
3471 queue_delayed_work(bond->wq, &bond->arp_work, 0);
Jay Vosburghf5b2b962006-09-22 21:54:53 -07003472 if (bond->params.arp_validate)
Jiri Pirko3aba8912011-04-19 03:48:16 +00003473 bond->recv_probe = bond_arp_rcv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474 }
3475
3476 if (bond->params.mode == BOND_MODE_8023AD) {
Adrian Bunka40745f2007-10-24 18:27:43 +02003477 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003478 queue_delayed_work(bond->wq, &bond->ad_work, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479 /* register to receive LACPDUs */
Jiri Pirko3aba8912011-04-19 03:48:16 +00003480 bond->recv_probe = bond_3ad_lacpdu_recv;
Jay Vosburghfd989c82008-11-04 17:51:16 -08003481 bond_3ad_initiate_agg_selection(bond, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003482 }
3483
3484 return 0;
3485}
3486
3487static int bond_close(struct net_device *bond_dev)
3488{
Wang Chen454d7c92008-11-12 23:37:49 -08003489 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 write_lock_bh(&bond->lock);
3492
Ben Hutchingsad246c92011-04-26 15:25:52 +00003493 bond->send_peer_notif = 0;
3494
Linus Torvalds1da177e2005-04-16 15:20:36 -07003495 /* signal timers not to re-arm */
3496 bond->kill_timers = 1;
3497
3498 write_unlock_bh(&bond->lock);
3499
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500 if (bond->params.miimon) { /* link check interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003501 cancel_delayed_work(&bond->mii_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502 }
3503
3504 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003505 cancel_delayed_work(&bond->arp_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003506 }
3507
3508 switch (bond->params.mode) {
3509 case BOND_MODE_8023AD:
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003510 cancel_delayed_work(&bond->ad_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511 break;
3512 case BOND_MODE_TLB:
3513 case BOND_MODE_ALB:
Jay Vosburgh1b76b312007-10-17 17:37:45 -07003514 cancel_delayed_work(&bond->alb_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515 break;
3516 default:
3517 break;
3518 }
3519
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00003520 if (delayed_work_pending(&bond->mcast_work))
3521 cancel_delayed_work(&bond->mcast_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522
Holger Eitzenberger58402052008-12-09 23:07:13 -08003523 if (bond_is_lb(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524 /* Must be called only after all
3525 * slaves have been released
3526 */
3527 bond_alb_deinitialize(bond);
3528 }
Jiri Pirko3aba8912011-04-19 03:48:16 +00003529 bond->recv_probe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530
3531 return 0;
3532}
3533
Eric Dumazet28172732010-07-07 14:58:56 -07003534static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3535 struct rtnl_link_stats64 *stats)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536{
Wang Chen454d7c92008-11-12 23:37:49 -08003537 struct bonding *bond = netdev_priv(bond_dev);
Eric Dumazet28172732010-07-07 14:58:56 -07003538 struct rtnl_link_stats64 temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539 struct slave *slave;
3540 int i;
3541
Eric Dumazet28172732010-07-07 14:58:56 -07003542 memset(stats, 0, sizeof(*stats));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543
3544 read_lock_bh(&bond->lock);
3545
3546 bond_for_each_slave(bond, slave, i) {
Ben Hutchingsbe1f3c22010-06-08 07:19:54 +00003547 const struct rtnl_link_stats64 *sstats =
Eric Dumazet28172732010-07-07 14:58:56 -07003548 dev_get_stats(slave->dev, &temp);
Stephen Hemmingereeda3fd2008-11-19 21:40:23 -08003549
Eric Dumazet28172732010-07-07 14:58:56 -07003550 stats->rx_packets += sstats->rx_packets;
3551 stats->rx_bytes += sstats->rx_bytes;
3552 stats->rx_errors += sstats->rx_errors;
3553 stats->rx_dropped += sstats->rx_dropped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554
Eric Dumazet28172732010-07-07 14:58:56 -07003555 stats->tx_packets += sstats->tx_packets;
3556 stats->tx_bytes += sstats->tx_bytes;
3557 stats->tx_errors += sstats->tx_errors;
3558 stats->tx_dropped += sstats->tx_dropped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559
Eric Dumazet28172732010-07-07 14:58:56 -07003560 stats->multicast += sstats->multicast;
3561 stats->collisions += sstats->collisions;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562
Eric Dumazet28172732010-07-07 14:58:56 -07003563 stats->rx_length_errors += sstats->rx_length_errors;
3564 stats->rx_over_errors += sstats->rx_over_errors;
3565 stats->rx_crc_errors += sstats->rx_crc_errors;
3566 stats->rx_frame_errors += sstats->rx_frame_errors;
3567 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3568 stats->rx_missed_errors += sstats->rx_missed_errors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569
Eric Dumazet28172732010-07-07 14:58:56 -07003570 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3571 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3572 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3573 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3574 stats->tx_window_errors += sstats->tx_window_errors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575 }
3576
3577 read_unlock_bh(&bond->lock);
3578
3579 return stats;
3580}
3581
3582static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3583{
3584 struct net_device *slave_dev = NULL;
3585 struct ifbond k_binfo;
3586 struct ifbond __user *u_binfo = NULL;
3587 struct ifslave k_sinfo;
3588 struct ifslave __user *u_sinfo = NULL;
3589 struct mii_ioctl_data *mii = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590 int res = 0;
3591
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003592 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593
3594 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595 case SIOCGMIIPHY:
3596 mii = if_mii(ifr);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003597 if (!mii)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598 return -EINVAL;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003599
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600 mii->phy_id = 0;
3601 /* Fall Through */
3602 case SIOCGMIIREG:
3603 /*
3604 * We do this again just in case we were called by SIOCGMIIREG
3605 * instead of SIOCGMIIPHY.
3606 */
3607 mii = if_mii(ifr);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003608 if (!mii)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609 return -EINVAL;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003610
Linus Torvalds1da177e2005-04-16 15:20:36 -07003611
3612 if (mii->reg_num == 1) {
Wang Chen454d7c92008-11-12 23:37:49 -08003613 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 mii->val_out = 0;
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07003615 read_lock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616 read_lock(&bond->curr_slave_lock);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003617 if (netif_carrier_ok(bond->dev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618 mii->val_out = BMSR_LSTATUS;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003619
Linus Torvalds1da177e2005-04-16 15:20:36 -07003620 read_unlock(&bond->curr_slave_lock);
Jay Vosburgh6603a6f2007-10-17 17:37:50 -07003621 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003622 }
3623
3624 return 0;
3625 case BOND_INFO_QUERY_OLD:
3626 case SIOCBONDINFOQUERY:
3627 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3628
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003629 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003630 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631
3632 res = bond_info_query(bond_dev, &k_binfo);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003633 if (res == 0 &&
3634 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3635 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636
3637 return res;
3638 case BOND_SLAVE_INFO_QUERY_OLD:
3639 case SIOCBONDSLAVEINFOQUERY:
3640 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3641
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003642 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644
3645 res = bond_slave_info_query(bond_dev, &k_sinfo);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003646 if (res == 0 &&
3647 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3648 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649
3650 return res;
3651 default:
3652 /* Go on */
3653 break;
3654 }
3655
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003656 if (!capable(CAP_NET_ADMIN))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00003659 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003661 pr_debug("slave_dev=%p:\n", slave_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003663 if (!slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664 res = -ENODEV;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003665 else {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003666 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667 switch (cmd) {
3668 case BOND_ENSLAVE_OLD:
3669 case SIOCBONDENSLAVE:
3670 res = bond_enslave(bond_dev, slave_dev);
3671 break;
3672 case BOND_RELEASE_OLD:
3673 case SIOCBONDRELEASE:
3674 res = bond_release(bond_dev, slave_dev);
3675 break;
3676 case BOND_SETHWADDR_OLD:
3677 case SIOCBONDSETHWADDR:
3678 res = bond_sethwaddr(bond_dev, slave_dev);
3679 break;
3680 case BOND_CHANGE_ACTIVE_OLD:
3681 case SIOCBONDCHANGEACTIVE:
3682 res = bond_ioctl_change_active(bond_dev, slave_dev);
3683 break;
3684 default:
3685 res = -EOPNOTSUPP;
3686 }
3687
3688 dev_put(slave_dev);
3689 }
3690
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691 return res;
3692}
3693
Jiri Pirko22bedad32010-04-01 21:22:57 +00003694static bool bond_addr_in_mc_list(unsigned char *addr,
3695 struct netdev_hw_addr_list *list,
3696 int addrlen)
3697{
3698 struct netdev_hw_addr *ha;
3699
3700 netdev_hw_addr_list_for_each(ha, list)
3701 if (!memcmp(ha->addr, addr, addrlen))
3702 return true;
3703
3704 return false;
3705}
3706
Jiri Pirkod03462b2011-08-16 03:15:04 +00003707static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3708{
3709 struct bonding *bond = netdev_priv(bond_dev);
3710
3711 if (change & IFF_PROMISC)
3712 bond_set_promiscuity(bond,
3713 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3714
3715 if (change & IFF_ALLMULTI)
3716 bond_set_allmulti(bond,
3717 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3718}
3719
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720static void bond_set_multicast_list(struct net_device *bond_dev)
3721{
Wang Chen454d7c92008-11-12 23:37:49 -08003722 struct bonding *bond = netdev_priv(bond_dev);
Jiri Pirko22bedad32010-04-01 21:22:57 +00003723 struct netdev_hw_addr *ha;
3724 bool found;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725
Jay Vosburgh80ee5ad2008-01-29 18:07:44 -08003726 read_lock(&bond->lock);
3727
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728 /* looking for addresses to add to slaves' mc list */
Jiri Pirko22bedad32010-04-01 21:22:57 +00003729 netdev_for_each_mc_addr(ha, bond_dev) {
3730 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3731 bond_dev->addr_len);
3732 if (!found)
3733 bond_mc_add(bond, ha->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734 }
3735
3736 /* looking for addresses to delete from slaves' list */
Jiri Pirko22bedad32010-04-01 21:22:57 +00003737 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3738 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3739 bond_dev->addr_len);
3740 if (!found)
3741 bond_mc_del(bond, ha->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 }
3743
3744 /* save master's multicast list */
Jiri Pirko22bedad32010-04-01 21:22:57 +00003745 __hw_addr_flush(&bond->mc_list);
3746 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3747 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748
Jay Vosburgh80ee5ad2008-01-29 18:07:44 -08003749 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003750}
3751
Stephen Hemminger00829822008-11-20 20:14:53 -08003752static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3753{
3754 struct bonding *bond = netdev_priv(dev);
3755 struct slave *slave = bond->first_slave;
3756
3757 if (slave) {
3758 const struct net_device_ops *slave_ops
3759 = slave->dev->netdev_ops;
3760 if (slave_ops->ndo_neigh_setup)
Patrick McHardy72e22402009-03-05 01:57:44 -08003761 return slave_ops->ndo_neigh_setup(slave->dev, parms);
Stephen Hemminger00829822008-11-20 20:14:53 -08003762 }
3763 return 0;
3764}
3765
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766/*
3767 * Change the MTU of all of a master's slaves to match the master
3768 */
3769static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3770{
Wang Chen454d7c92008-11-12 23:37:49 -08003771 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772 struct slave *slave, *stop_at;
3773 int res = 0;
3774 int i;
3775
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003776 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003777 (bond_dev ? bond_dev->name : "None"), new_mtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778
3779 /* Can't hold bond->lock with bh disabled here since
3780 * some base drivers panic. On the other hand we can't
3781 * hold bond->lock without bh disabled because we'll
3782 * deadlock. The only solution is to rely on the fact
3783 * that we're under rtnl_lock here, and the slaves
3784 * list won't change. This doesn't solve the problem
3785 * of setting the slave's MTU while it is
3786 * transmitting, but the assumption is that the base
3787 * driver can handle that.
3788 *
3789 * TODO: figure out a way to safely iterate the slaves
3790 * list, but without holding a lock around the actual
3791 * call to the base driver.
3792 */
3793
3794 bond_for_each_slave(bond, slave, i) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003795 pr_debug("s %p s->p %p c_m %p\n",
3796 slave,
3797 slave->prev,
3798 slave->dev->netdev_ops->ndo_change_mtu);
Mitch Williamse944ef72005-11-09 10:36:50 -08003799
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800 res = dev_set_mtu(slave->dev, new_mtu);
3801
3802 if (res) {
3803 /* If we failed to set the slave's mtu to the new value
3804 * we must abort the operation even in ACTIVE_BACKUP
3805 * mode, because if we allow the backup slaves to have
3806 * different mtu values than the active slave we'll
3807 * need to change their mtu when doing a failover. That
3808 * means changing their mtu from timer context, which
3809 * is probably not a good idea.
3810 */
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003811 pr_debug("err %d %s\n", res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 goto unwind;
3813 }
3814 }
3815
3816 bond_dev->mtu = new_mtu;
3817
3818 return 0;
3819
3820unwind:
3821 /* unwind from head to the slave that failed */
3822 stop_at = slave;
3823 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3824 int tmp_res;
3825
3826 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3827 if (tmp_res) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003828 pr_debug("unwind err %d dev %s\n",
3829 tmp_res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830 }
3831 }
3832
3833 return res;
3834}
3835
3836/*
3837 * Change HW address
3838 *
3839 * Note that many devices must be down to change the HW address, and
3840 * downing the master releases all slaves. We can make bonds full of
3841 * bonding devices to test this, however.
3842 */
3843static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3844{
Wang Chen454d7c92008-11-12 23:37:49 -08003845 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846 struct sockaddr *sa = addr, tmp_sa;
3847 struct slave *slave, *stop_at;
3848 int res = 0;
3849 int i;
3850
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003851 if (bond->params.mode == BOND_MODE_ALB)
3852 return bond_alb_set_mac_address(bond_dev, addr);
3853
3854
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003855 pr_debug("bond=%p, name=%s\n",
3856 bond, bond_dev ? bond_dev->name : "None");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857
Jay Vosburghdd957c52007-10-09 19:57:24 -07003858 /*
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07003859 * If fail_over_mac is set to active, do nothing and return
3860 * success. Returning an error causes ifenslave to fail.
Jay Vosburghdd957c52007-10-09 19:57:24 -07003861 */
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07003862 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
Jay Vosburghdd957c52007-10-09 19:57:24 -07003863 return 0;
Moni Shoua2ab82852007-10-09 19:43:39 -07003864
Stephen Hemminger3d632c32009-06-12 19:02:48 +00003865 if (!is_valid_ether_addr(sa->sa_data))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866 return -EADDRNOTAVAIL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867
3868 /* Can't hold bond->lock with bh disabled here since
3869 * some base drivers panic. On the other hand we can't
3870 * hold bond->lock without bh disabled because we'll
3871 * deadlock. The only solution is to rely on the fact
3872 * that we're under rtnl_lock here, and the slaves
3873 * list won't change. This doesn't solve the problem
3874 * of setting the slave's hw address while it is
3875 * transmitting, but the assumption is that the base
3876 * driver can handle that.
3877 *
3878 * TODO: figure out a way to safely iterate the slaves
3879 * list, but without holding a lock around the actual
3880 * call to the base driver.
3881 */
3882
3883 bond_for_each_slave(bond, slave, i) {
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003884 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003885 pr_debug("slave %p %s\n", slave, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003886
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08003887 if (slave_ops->ndo_set_mac_address == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888 res = -EOPNOTSUPP;
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003889 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890 goto unwind;
3891 }
3892
3893 res = dev_set_mac_address(slave->dev, addr);
3894 if (res) {
3895 /* TODO: consider downing the slave
3896 * and retry ?
3897 * User should expect communications
3898 * breakage anyway until ARP finish
3899 * updating, so...
3900 */
Holger Eitzenberger5a03cdb2008-12-09 23:09:22 -08003901 pr_debug("err %d %s\n", res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902 goto unwind;
3903 }
3904 }
3905
3906 /* success */
3907 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3908 return 0;
3909
3910unwind:
3911 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3912 tmp_sa.sa_family = bond_dev->type;
3913
3914 /* unwind from head to the slave that failed */
3915 stop_at = slave;
3916 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3917 int tmp_res;
3918
3919 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3920 if (tmp_res) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08003921 pr_debug("unwind err %d dev %s\n",
3922 tmp_res, slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923 }
3924 }
3925
3926 return res;
3927}
3928
3929static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3930{
Wang Chen454d7c92008-11-12 23:37:49 -08003931 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932 struct slave *slave, *start_at;
Jay Vosburghcf5f9042007-10-17 17:37:47 -07003933 int i, slave_no, res = 1;
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003934 struct iphdr *iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003935
Jay Vosburghcf5f9042007-10-17 17:37:47 -07003936 /*
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003937 * Start with the curr_active_slave that joined the bond as the
3938 * default for sending IGMP traffic. For failover purposes one
3939 * needs to maintain some consistency for the interface that will
3940 * send the join/membership reports. The curr_active_slave found
3941 * will send all of this type of traffic.
Jay Vosburghcf5f9042007-10-17 17:37:47 -07003942 */
Eric Dumazet00ae7022010-03-30 23:08:37 +00003943 if ((iph->protocol == IPPROTO_IGMP) &&
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003944 (skb->protocol == htons(ETH_P_IP))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003945
Andy Gospodareka2fd9402010-03-25 14:49:05 +00003946 read_lock(&bond->curr_slave_lock);
3947 slave = bond->curr_active_slave;
3948 read_unlock(&bond->curr_slave_lock);
3949
3950 if (!slave)
3951 goto out;
3952 } else {
3953 /*
3954 * Concurrent TX may collide on rr_tx_counter; we accept
3955 * that as being rare enough not to justify using an
3956 * atomic op here.
3957 */
3958 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3959
3960 bond_for_each_slave(bond, slave, i) {
3961 slave_no--;
3962 if (slave_no < 0)
3963 break;
3964 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 }
3966
Jay Vosburghcf5f9042007-10-17 17:37:47 -07003967 start_at = slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968 bond_for_each_slave_from(bond, slave, i, start_at) {
3969 if (IS_UP(slave->dev) &&
3970 (slave->link == BOND_LINK_UP) &&
Jiri Pirkoe30bc062011-03-12 03:14:37 +00003971 bond_is_active_slave(slave)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 res = bond_dev_queue_xmit(bond, skb, slave->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973 break;
3974 }
3975 }
3976
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977out:
3978 if (res) {
3979 /* no suitable interface, frame not sent */
3980 dev_kfree_skb(skb);
3981 }
Michał Mirosław0693e882011-05-07 01:48:02 +00003982
Patrick McHardyec634fe2009-07-05 19:23:38 -07003983 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984}
3985
John W. Linville075897c2005-09-28 17:50:53 -04003986
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987/*
3988 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3989 * the bond has a usable interface.
3990 */
3991static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3992{
Wang Chen454d7c92008-11-12 23:37:49 -08003993 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994 int res = 1;
3995
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996 read_lock(&bond->curr_slave_lock);
3997
Michał Mirosław0693e882011-05-07 01:48:02 +00003998 if (bond->curr_active_slave)
3999 res = bond_dev_queue_xmit(bond, skb,
4000 bond->curr_active_slave->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004002 if (res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003 /* no suitable interface, frame not sent */
4004 dev_kfree_skb(skb);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004005
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006 read_unlock(&bond->curr_slave_lock);
Michał Mirosław0693e882011-05-07 01:48:02 +00004007
Patrick McHardyec634fe2009-07-05 19:23:38 -07004008 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009}
4010
4011/*
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004012 * In bond_xmit_xor() , we determine the output device by using a pre-
4013 * determined xmit_hash_policy(), If the selected device is not enabled,
4014 * find the next active slave.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015 */
4016static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4017{
Wang Chen454d7c92008-11-12 23:37:49 -08004018 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019 struct slave *slave, *start_at;
4020 int slave_no;
4021 int i;
4022 int res = 1;
4023
Jasper Spaansa361c832009-10-23 04:09:24 +00004024 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025
4026 bond_for_each_slave(bond, slave, i) {
4027 slave_no--;
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004028 if (slave_no < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030 }
4031
4032 start_at = slave;
4033
4034 bond_for_each_slave_from(bond, slave, i, start_at) {
4035 if (IS_UP(slave->dev) &&
4036 (slave->link == BOND_LINK_UP) &&
Jiri Pirkoe30bc062011-03-12 03:14:37 +00004037 bond_is_active_slave(slave)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4039 break;
4040 }
4041 }
4042
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043 if (res) {
4044 /* no suitable interface, frame not sent */
4045 dev_kfree_skb(skb);
4046 }
Michał Mirosław0693e882011-05-07 01:48:02 +00004047
Patrick McHardyec634fe2009-07-05 19:23:38 -07004048 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004049}
4050
4051/*
4052 * in broadcast mode, we send everything to all usable interfaces.
4053 */
4054static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4055{
Wang Chen454d7c92008-11-12 23:37:49 -08004056 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004057 struct slave *slave, *start_at;
4058 struct net_device *tx_dev = NULL;
4059 int i;
4060 int res = 1;
4061
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062 read_lock(&bond->curr_slave_lock);
4063 start_at = bond->curr_active_slave;
4064 read_unlock(&bond->curr_slave_lock);
4065
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004066 if (!start_at)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004068
4069 bond_for_each_slave_from(bond, slave, i, start_at) {
4070 if (IS_UP(slave->dev) &&
4071 (slave->link == BOND_LINK_UP) &&
Jiri Pirkoe30bc062011-03-12 03:14:37 +00004072 bond_is_active_slave(slave)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073 if (tx_dev) {
4074 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4075 if (!skb2) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004076 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
Mitch Williams4e0952c2005-11-09 10:34:57 -08004077 bond_dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004078 continue;
4079 }
4080
4081 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4082 if (res) {
4083 dev_kfree_skb(skb2);
4084 continue;
4085 }
4086 }
4087 tx_dev = slave->dev;
4088 }
4089 }
4090
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004091 if (tx_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004092 res = bond_dev_queue_xmit(bond, skb, tx_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004093
4094out:
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004095 if (res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004096 /* no suitable interface, frame not sent */
4097 dev_kfree_skb(skb);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004098
Linus Torvalds1da177e2005-04-16 15:20:36 -07004099 /* frame sent to all suitable interfaces */
Patrick McHardyec634fe2009-07-05 19:23:38 -07004100 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004101}
4102
4103/*------------------------- Device initialization ---------------------------*/
4104
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08004105static void bond_set_xmit_hash_policy(struct bonding *bond)
4106{
4107 switch (bond->params.xmit_policy) {
4108 case BOND_XMIT_POLICY_LAYER23:
4109 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4110 break;
4111 case BOND_XMIT_POLICY_LAYER34:
4112 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4113 break;
4114 case BOND_XMIT_POLICY_LAYER2:
4115 default:
4116 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4117 break;
4118 }
4119}
4120
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004121/*
4122 * Lookup the slave that corresponds to a qid
4123 */
4124static inline int bond_slave_override(struct bonding *bond,
4125 struct sk_buff *skb)
4126{
4127 int i, res = 1;
4128 struct slave *slave = NULL;
4129 struct slave *check_slave;
4130
Michał Mirosław0693e882011-05-07 01:48:02 +00004131 if (!skb->queue_mapping)
4132 return 1;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004133
4134 /* Find out if any slaves have the same mapping as this skb. */
4135 bond_for_each_slave(bond, check_slave, i) {
4136 if (check_slave->queue_id == skb->queue_mapping) {
4137 slave = check_slave;
4138 break;
4139 }
4140 }
4141
4142 /* If the slave isn't UP, use default transmit policy. */
4143 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4144 (slave->link == BOND_LINK_UP)) {
4145 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4146 }
4147
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004148 return res;
4149}
4150
Neil Horman374eeb52011-06-03 10:35:52 +00004151
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004152static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4153{
4154 /*
4155 * This helper function exists to help dev_pick_tx get the correct
Phil Oesterfd0e4352011-03-14 06:22:04 +00004156 * destination queue. Using a helper function skips a call to
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004157 * skb_tx_hash and will put the skbs in the queue we expect on their
4158 * way down to the bonding driver.
4159 */
Phil Oesterfd0e4352011-03-14 06:22:04 +00004160 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4161
Neil Horman374eeb52011-06-03 10:35:52 +00004162 /*
4163 * Save the original txq to restore before passing to the driver
4164 */
4165 bond_queue_mapping(skb) = skb->queue_mapping;
4166
Phil Oesterfd0e4352011-03-14 06:22:04 +00004167 if (unlikely(txq >= dev->real_num_tx_queues)) {
David Decotignyd30ee672011-04-13 15:22:29 +00004168 do {
Phil Oesterfd0e4352011-03-14 06:22:04 +00004169 txq -= dev->real_num_tx_queues;
David Decotignyd30ee672011-04-13 15:22:29 +00004170 } while (txq >= dev->real_num_tx_queues);
Phil Oesterfd0e4352011-03-14 06:22:04 +00004171 }
4172 return txq;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004173}
4174
Michał Mirosław0693e882011-05-07 01:48:02 +00004175static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
Stephen Hemminger00829822008-11-20 20:14:53 -08004176{
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004177 struct bonding *bond = netdev_priv(dev);
4178
4179 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4180 if (!bond_slave_override(bond, skb))
4181 return NETDEV_TX_OK;
4182 }
Stephen Hemminger00829822008-11-20 20:14:53 -08004183
4184 switch (bond->params.mode) {
4185 case BOND_MODE_ROUNDROBIN:
4186 return bond_xmit_roundrobin(skb, dev);
4187 case BOND_MODE_ACTIVEBACKUP:
4188 return bond_xmit_activebackup(skb, dev);
4189 case BOND_MODE_XOR:
4190 return bond_xmit_xor(skb, dev);
4191 case BOND_MODE_BROADCAST:
4192 return bond_xmit_broadcast(skb, dev);
4193 case BOND_MODE_8023AD:
4194 return bond_3ad_xmit_xor(skb, dev);
4195 case BOND_MODE_ALB:
4196 case BOND_MODE_TLB:
4197 return bond_alb_xmit(skb, dev);
4198 default:
4199 /* Should never happen, mode already checked */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004200 pr_err("%s: Error: Unknown bonding mode %d\n",
4201 dev->name, bond->params.mode);
Stephen Hemminger00829822008-11-20 20:14:53 -08004202 WARN_ON_ONCE(1);
4203 dev_kfree_skb(skb);
4204 return NETDEV_TX_OK;
4205 }
4206}
4207
Michał Mirosław0693e882011-05-07 01:48:02 +00004208static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4209{
4210 struct bonding *bond = netdev_priv(dev);
4211 netdev_tx_t ret = NETDEV_TX_OK;
4212
4213 /*
4214 * If we risk deadlock from transmitting this in the
4215 * netpoll path, tell netpoll to queue the frame for later tx
4216 */
4217 if (is_netpoll_tx_blocked(dev))
4218 return NETDEV_TX_BUSY;
4219
4220 read_lock(&bond->lock);
4221
4222 if (bond->slave_cnt)
4223 ret = __bond_start_xmit(skb, dev);
4224 else
4225 dev_kfree_skb(skb);
4226
4227 read_unlock(&bond->lock);
4228
4229 return ret;
4230}
Stephen Hemminger00829822008-11-20 20:14:53 -08004231
Linus Torvalds1da177e2005-04-16 15:20:36 -07004232/*
4233 * set bond mode specific net device operations
4234 */
Mitch Williamsa77b5322005-11-09 10:35:51 -08004235void bond_set_mode_ops(struct bonding *bond, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004236{
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004237 struct net_device *bond_dev = bond->dev;
4238
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239 switch (mode) {
4240 case BOND_MODE_ROUNDROBIN:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004241 break;
4242 case BOND_MODE_ACTIVEBACKUP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243 break;
4244 case BOND_MODE_XOR:
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08004245 bond_set_xmit_hash_policy(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004246 break;
4247 case BOND_MODE_BROADCAST:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248 break;
4249 case BOND_MODE_8023AD:
Jay Vosburgh6f6652b2007-12-06 23:40:34 -08004250 bond_set_xmit_hash_policy(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004251 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004252 case BOND_MODE_ALB:
Jay Vosburgh8f903c72006-02-21 16:36:44 -08004253 /* FALLTHRU */
4254 case BOND_MODE_TLB:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004255 break;
4256 default:
4257 /* Should never happen, mode already checked */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004258 pr_err("%s: Error: Unknown bonding mode %d\n",
4259 bond_dev->name, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260 break;
4261 }
4262}
4263
Jay Vosburgh217df672005-09-26 16:11:50 -07004264static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4265 struct ethtool_drvinfo *drvinfo)
4266{
4267 strncpy(drvinfo->driver, DRV_NAME, 32);
4268 strncpy(drvinfo->version, DRV_VERSION, 32);
4269 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4270}
4271
Jeff Garzik7282d492006-09-13 14:30:00 -04004272static const struct ethtool_ops bond_ethtool_ops = {
Jay Vosburgh217df672005-09-26 16:11:50 -07004273 .get_drvinfo = bond_ethtool_get_drvinfo,
Stephen Hemmingerfa53eba2008-09-13 21:17:09 -04004274 .get_link = ethtool_op_get_link,
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004275};
4276
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004277static const struct net_device_ops bond_netdev_ops = {
Stephen Hemminger181470f2009-06-12 19:02:52 +00004278 .ndo_init = bond_init,
Stephen Hemminger9e716262009-06-12 19:02:47 +00004279 .ndo_uninit = bond_uninit,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004280 .ndo_open = bond_open,
4281 .ndo_stop = bond_close,
Stephen Hemminger00829822008-11-20 20:14:53 -08004282 .ndo_start_xmit = bond_start_xmit,
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004283 .ndo_select_queue = bond_select_queue,
Ben Hutchingsbe1f3c22010-06-08 07:19:54 +00004284 .ndo_get_stats64 = bond_get_stats,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004285 .ndo_do_ioctl = bond_do_ioctl,
Jiri Pirkod03462b2011-08-16 03:15:04 +00004286 .ndo_change_rx_flags = bond_change_rx_flags,
Jiri Pirkoafc4b132011-08-16 06:29:01 +00004287 .ndo_set_rx_mode = bond_set_multicast_list,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004288 .ndo_change_mtu = bond_change_mtu,
Jiri Pirkocc0e4072011-07-20 04:54:46 +00004289 .ndo_set_mac_address = bond_set_mac_address,
Stephen Hemminger00829822008-11-20 20:14:53 -08004290 .ndo_neigh_setup = bond_neigh_setup,
Jiri Pirkocc0e4072011-07-20 04:54:46 +00004291 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004292 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
WANG Congf6dc31a2010-05-06 00:48:51 -07004293#ifdef CONFIG_NET_POLL_CONTROLLER
Amerigo Wang8a8efa22011-02-17 23:43:32 +00004294 .ndo_netpoll_setup = bond_netpoll_setup,
WANG Congf6dc31a2010-05-06 00:48:51 -07004295 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4296 .ndo_poll_controller = bond_poll_controller,
4297#endif
Jiri Pirko9232ecc2011-02-13 09:33:01 +00004298 .ndo_add_slave = bond_enslave,
4299 .ndo_del_slave = bond_release,
Michał Mirosławb2a103e2011-05-07 03:22:17 +00004300 .ndo_fix_features = bond_fix_features,
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004301};
4302
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004303static void bond_destructor(struct net_device *bond_dev)
4304{
4305 struct bonding *bond = netdev_priv(bond_dev);
4306 if (bond->wq)
4307 destroy_workqueue(bond->wq);
4308 free_netdev(bond_dev);
4309}
4310
Stephen Hemminger181470f2009-06-12 19:02:52 +00004311static void bond_setup(struct net_device *bond_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004312{
Wang Chen454d7c92008-11-12 23:37:49 -08004313 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004314
Linus Torvalds1da177e2005-04-16 15:20:36 -07004315 /* initialize rwlocks */
4316 rwlock_init(&bond->lock);
4317 rwlock_init(&bond->curr_slave_lock);
4318
Stephen Hemmingerd2991f72009-06-12 19:02:44 +00004319 bond->params = bonding_defaults;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004320
4321 /* Initialize pointers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004322 bond->dev = bond_dev;
4323 INIT_LIST_HEAD(&bond->vlan_list);
4324
4325 /* Initialize the device entry points */
Stephen Hemminger181470f2009-06-12 19:02:52 +00004326 ether_setup(bond_dev);
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08004327 bond_dev->netdev_ops = &bond_netdev_ops;
Arthur Kepner8531c5f2005-08-23 01:34:53 -04004328 bond_dev->ethtool_ops = &bond_ethtool_ops;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004329 bond_set_mode_ops(bond, bond->params.mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004330
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004331 bond_dev->destructor = bond_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004332
4333 /* Initialize the device options */
4334 bond_dev->tx_queue_len = 0;
4335 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
Jay Vosburgh0b680e72006-09-22 21:54:10 -07004336 bond_dev->priv_flags |= IFF_BONDING;
Neil Horman550fd082011-07-26 06:05:38 +00004337 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004338
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339 /* At first, we block adding VLANs. That's the only way to
4340 * prevent problems that occur when adding VLANs over an
4341 * empty bond. The block will be removed once non-challenged
4342 * slaves are enslaved.
4343 */
4344 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4345
Herbert Xu932ff272006-06-09 12:20:56 -07004346 /* don't acquire bond device's netif_tx_lock when
Linus Torvalds1da177e2005-04-16 15:20:36 -07004347 * transmitting */
4348 bond_dev->features |= NETIF_F_LLTX;
4349
4350 /* By default, we declare the bond to be fully
4351 * VLAN hardware accelerated capable. Special
4352 * care is taken in the various xmit functions
4353 * when there are slaves that are not hw accel
4354 * capable
4355 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004356
Michał Mirosławb2a103e2011-05-07 03:22:17 +00004357 bond_dev->hw_features = BOND_VLAN_FEATURES |
4358 NETIF_F_HW_VLAN_TX |
4359 NETIF_F_HW_VLAN_RX |
4360 NETIF_F_HW_VLAN_FILTER;
4361
4362 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4363 bond_dev->features |= bond_dev->hw_features;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004364}
4365
Jay Vosburghfdaea7a2007-12-06 23:40:35 -08004366static void bond_work_cancel_all(struct bonding *bond)
4367{
4368 write_lock_bh(&bond->lock);
4369 bond->kill_timers = 1;
4370 write_unlock_bh(&bond->lock);
4371
4372 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4373 cancel_delayed_work(&bond->mii_work);
4374
4375 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4376 cancel_delayed_work(&bond->arp_work);
4377
4378 if (bond->params.mode == BOND_MODE_ALB &&
4379 delayed_work_pending(&bond->alb_work))
4380 cancel_delayed_work(&bond->alb_work);
4381
4382 if (bond->params.mode == BOND_MODE_8023AD &&
4383 delayed_work_pending(&bond->ad_work))
4384 cancel_delayed_work(&bond->ad_work);
Flavio Leitner5a37e8c2010-10-05 14:23:57 +00004385
4386 if (delayed_work_pending(&bond->mcast_work))
4387 cancel_delayed_work(&bond->mcast_work);
Jay Vosburghfdaea7a2007-12-06 23:40:35 -08004388}
4389
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004390/*
4391* Destroy a bonding device.
4392* Must be under rtnl_lock when this function is called.
4393*/
4394static void bond_uninit(struct net_device *bond_dev)
Jay Vosburgha434e432008-10-30 17:41:15 -07004395{
Wang Chen454d7c92008-11-12 23:37:49 -08004396 struct bonding *bond = netdev_priv(bond_dev);
Jay Vosburghf35188f2010-07-21 12:14:47 +00004397 struct vlan_entry *vlan, *tmp;
Jay Vosburgha434e432008-10-30 17:41:15 -07004398
WANG Congf6dc31a2010-05-06 00:48:51 -07004399 bond_netpoll_cleanup(bond_dev);
4400
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004401 /* Release the bonded slaves */
4402 bond_release_all(bond_dev);
4403
Jay Vosburgha434e432008-10-30 17:41:15 -07004404 list_del(&bond->bond_list);
4405
4406 bond_work_cancel_all(bond);
4407
Jay Vosburgha434e432008-10-30 17:41:15 -07004408 bond_remove_proc_entry(bond);
Eric W. Biedermanc67dfb22009-10-29 14:18:24 +00004409
Taku Izumif073c7c2010-12-09 15:17:13 +00004410 bond_debug_unregister(bond);
4411
Jiri Pirko22bedad32010-04-01 21:22:57 +00004412 __hw_addr_flush(&bond->mc_list);
Jay Vosburghf35188f2010-07-21 12:14:47 +00004413
4414 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4415 list_del(&vlan->vlan_list);
4416 kfree(vlan);
4417 }
Jay Vosburgha434e432008-10-30 17:41:15 -07004418}
4419
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420/*------------------------- Module initialization ---------------------------*/
4421
4422/*
4423 * Convert string input module parms. Accept either the
Jay Vosburghece95f72008-01-17 16:25:01 -08004424 * number of the mode or its string name. A bit complicated because
4425 * some mode names are substrings of other names, and calls from sysfs
4426 * may have whitespace in the name (trailing newlines, for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07004427 */
Holger Eitzenberger325dcf72008-12-09 23:10:17 -08004428int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004429{
Hannes Eder54b87322009-02-14 11:15:49 +00004430 int modeint = -1, i, rv;
Jay Vosburgha42e5342008-01-29 18:07:43 -08004431 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
Jay Vosburghece95f72008-01-17 16:25:01 -08004432
Jay Vosburgha42e5342008-01-29 18:07:43 -08004433 for (p = (char *)buf; *p; p++)
4434 if (!(isdigit(*p) || isspace(*p)))
4435 break;
4436
4437 if (*p)
Jay Vosburghece95f72008-01-17 16:25:01 -08004438 rv = sscanf(buf, "%20s", modestr);
Jay Vosburgha42e5342008-01-29 18:07:43 -08004439 else
Hannes Eder54b87322009-02-14 11:15:49 +00004440 rv = sscanf(buf, "%d", &modeint);
Jay Vosburgha42e5342008-01-29 18:07:43 -08004441
4442 if (!rv)
4443 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004444
4445 for (i = 0; tbl[i].modename; i++) {
Hannes Eder54b87322009-02-14 11:15:49 +00004446 if (modeint == tbl[i].mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447 return tbl[i].mode;
Jay Vosburghece95f72008-01-17 16:25:01 -08004448 if (strcmp(modestr, tbl[i].modename) == 0)
4449 return tbl[i].mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004450 }
4451
4452 return -1;
4453}
4454
4455static int bond_check_params(struct bond_params *params)
4456{
Jiri Pirkoa5499522009-09-25 03:28:09 +00004457 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004458
Linus Torvalds1da177e2005-04-16 15:20:36 -07004459 /*
4460 * Convert string parameters.
4461 */
4462 if (mode) {
4463 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4464 if (bond_mode == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004465 pr_err("Error: Invalid bonding mode \"%s\"\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004466 mode == NULL ? "NULL" : mode);
4467 return -EINVAL;
4468 }
4469 }
4470
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004471 if (xmit_hash_policy) {
4472 if ((bond_mode != BOND_MODE_XOR) &&
4473 (bond_mode != BOND_MODE_8023AD)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004474 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004475 bond_mode_name(bond_mode));
4476 } else {
4477 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4478 xmit_hashtype_tbl);
4479 if (xmit_hashtype == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004480 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004481 xmit_hash_policy == NULL ? "NULL" :
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004482 xmit_hash_policy);
4483 return -EINVAL;
4484 }
4485 }
4486 }
4487
Linus Torvalds1da177e2005-04-16 15:20:36 -07004488 if (lacp_rate) {
4489 if (bond_mode != BOND_MODE_8023AD) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004490 pr_info("lacp_rate param is irrelevant in mode %s\n",
4491 bond_mode_name(bond_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004492 } else {
4493 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4494 if (lacp_fast == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004495 pr_err("Error: Invalid lacp rate \"%s\"\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004496 lacp_rate == NULL ? "NULL" : lacp_rate);
4497 return -EINVAL;
4498 }
4499 }
4500 }
4501
Jay Vosburghfd989c82008-11-04 17:51:16 -08004502 if (ad_select) {
4503 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4504 if (params->ad_select == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004505 pr_err("Error: Invalid ad_select \"%s\"\n",
Jay Vosburghfd989c82008-11-04 17:51:16 -08004506 ad_select == NULL ? "NULL" : ad_select);
4507 return -EINVAL;
4508 }
4509
4510 if (bond_mode != BOND_MODE_8023AD) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004511 pr_warning("ad_select param only affects 802.3ad mode\n");
Jay Vosburghfd989c82008-11-04 17:51:16 -08004512 }
4513 } else {
4514 params->ad_select = BOND_AD_STABLE;
4515 }
4516
Nicolas de Pesloüanf5841302009-08-28 13:18:34 +00004517 if (max_bonds < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004518 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4519 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004520 max_bonds = BOND_DEFAULT_MAX_BONDS;
4521 }
4522
4523 if (miimon < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004524 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4525 miimon, INT_MAX, BOND_LINK_MON_INTERV);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004526 miimon = BOND_LINK_MON_INTERV;
4527 }
4528
4529 if (updelay < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004530 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4531 updelay, INT_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004532 updelay = 0;
4533 }
4534
4535 if (downdelay < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004536 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4537 downdelay, INT_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004538 downdelay = 0;
4539 }
4540
4541 if ((use_carrier != 0) && (use_carrier != 1)) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004542 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4543 use_carrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004544 use_carrier = 1;
4545 }
4546
Ben Hutchingsad246c92011-04-26 15:25:52 +00004547 if (num_peer_notif < 0 || num_peer_notif > 255) {
4548 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4549 num_peer_notif);
4550 num_peer_notif = 1;
4551 }
4552
Linus Torvalds1da177e2005-04-16 15:20:36 -07004553 /* reset values for 802.3ad */
4554 if (bond_mode == BOND_MODE_8023AD) {
4555 if (!miimon) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004556 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004557 pr_warning("Forcing miimon to 100msec\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004558 miimon = 100;
4559 }
4560 }
4561
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004562 if (tx_queues < 1 || tx_queues > 255) {
4563 pr_warning("Warning: tx_queues (%d) should be between "
4564 "1 and 255, resetting to %d\n",
4565 tx_queues, BOND_DEFAULT_TX_QUEUES);
4566 tx_queues = BOND_DEFAULT_TX_QUEUES;
4567 }
4568
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004569 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4570 pr_warning("Warning: all_slaves_active module parameter (%d), "
4571 "not of valid value (0/1), so it was set to "
4572 "0\n", all_slaves_active);
4573 all_slaves_active = 0;
4574 }
4575
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004576 if (resend_igmp < 0 || resend_igmp > 255) {
4577 pr_warning("Warning: resend_igmp (%d) should be between "
4578 "0 and 255, resetting to %d\n",
4579 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4580 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4581 }
4582
Linus Torvalds1da177e2005-04-16 15:20:36 -07004583 /* reset values for TLB/ALB */
4584 if ((bond_mode == BOND_MODE_TLB) ||
4585 (bond_mode == BOND_MODE_ALB)) {
4586 if (!miimon) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004587 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004588 pr_warning("Forcing miimon to 100msec\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004589 miimon = 100;
4590 }
4591 }
4592
4593 if (bond_mode == BOND_MODE_ALB) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004594 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4595 updelay);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004596 }
4597
4598 if (!miimon) {
4599 if (updelay || downdelay) {
4600 /* just warn the user the up/down delay will have
4601 * no effect since miimon is zero...
4602 */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004603 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4604 updelay, downdelay);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004605 }
4606 } else {
4607 /* don't allow arp monitoring */
4608 if (arp_interval) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004609 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4610 miimon, arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004611 arp_interval = 0;
4612 }
4613
4614 if ((updelay % miimon) != 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004615 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4616 updelay, miimon,
4617 (updelay / miimon) * miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004618 }
4619
4620 updelay /= miimon;
4621
4622 if ((downdelay % miimon) != 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004623 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4624 downdelay, miimon,
4625 (downdelay / miimon) * miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004626 }
4627
4628 downdelay /= miimon;
4629 }
4630
4631 if (arp_interval < 0) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004632 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4633 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004634 arp_interval = BOND_LINK_ARP_INTERV;
4635 }
4636
4637 for (arp_ip_count = 0;
4638 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4639 arp_ip_count++) {
4640 /* not complete check, but should be good enough to
4641 catch mistakes */
4642 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004643 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4644 arp_ip_target[arp_ip_count]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004645 arp_interval = 0;
4646 } else {
Al Virod3bb52b2007-08-22 20:06:58 -04004647 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004648 arp_target[arp_ip_count] = ip;
4649 }
4650 }
4651
4652 if (arp_interval && !arp_ip_count) {
4653 /* don't allow arping if no arp_ip_target given... */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004654 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4655 arp_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004656 arp_interval = 0;
4657 }
4658
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004659 if (arp_validate) {
4660 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004661 pr_err("arp_validate only supported in active-backup mode\n");
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004662 return -EINVAL;
4663 }
4664 if (!arp_interval) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004665 pr_err("arp_validate requires arp_interval\n");
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004666 return -EINVAL;
4667 }
4668
4669 arp_validate_value = bond_parse_parm(arp_validate,
4670 arp_validate_tbl);
4671 if (arp_validate_value == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004672 pr_err("Error: invalid arp_validate \"%s\"\n",
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004673 arp_validate == NULL ? "NULL" : arp_validate);
4674 return -EINVAL;
4675 }
4676 } else
4677 arp_validate_value = 0;
4678
Linus Torvalds1da177e2005-04-16 15:20:36 -07004679 if (miimon) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004680 pr_info("MII link monitoring set to %d ms\n", miimon);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004681 } else if (arp_interval) {
4682 int i;
4683
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004684 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4685 arp_interval,
4686 arp_validate_tbl[arp_validate_value].modename,
4687 arp_ip_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004688
4689 for (i = 0; i < arp_ip_count; i++)
Jiri Pirkoe5e2a8f2009-08-13 04:11:52 +00004690 pr_info(" %s", arp_ip_target[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004691
Jiri Pirkoe5e2a8f2009-08-13 04:11:52 +00004692 pr_info("\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004693
Jay Vosburghb8a97872008-06-13 18:12:04 -07004694 } else if (max_bonds) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004695 /* miimon and arp_interval not set, we need one so things
4696 * work as expected, see bonding.txt for details
4697 */
Andy Gospodarekb2730f42011-07-27 10:09:26 +00004698 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004699 }
4700
4701 if (primary && !USES_PRIMARY(bond_mode)) {
4702 /* currently, using a primary only makes sense
4703 * in active backup, TLB or ALB modes
4704 */
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004705 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4706 primary, bond_mode_name(bond_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004707 primary = NULL;
4708 }
4709
Jiri Pirkoa5499522009-09-25 03:28:09 +00004710 if (primary && primary_reselect) {
4711 primary_reselect_value = bond_parse_parm(primary_reselect,
4712 pri_reselect_tbl);
4713 if (primary_reselect_value == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004714 pr_err("Error: Invalid primary_reselect \"%s\"\n",
Jiri Pirkoa5499522009-09-25 03:28:09 +00004715 primary_reselect ==
4716 NULL ? "NULL" : primary_reselect);
4717 return -EINVAL;
4718 }
4719 } else {
4720 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4721 }
4722
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004723 if (fail_over_mac) {
4724 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4725 fail_over_mac_tbl);
4726 if (fail_over_mac_value == -1) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004727 pr_err("Error: invalid fail_over_mac \"%s\"\n",
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004728 arp_validate == NULL ? "NULL" : arp_validate);
4729 return -EINVAL;
4730 }
4731
4732 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004733 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004734 } else {
4735 fail_over_mac_value = BOND_FOM_NONE;
4736 }
Jay Vosburghdd957c52007-10-09 19:57:24 -07004737
Linus Torvalds1da177e2005-04-16 15:20:36 -07004738 /* fill params struct with the proper values */
4739 params->mode = bond_mode;
Jay Vosburgh169a3e62005-06-26 17:54:11 -04004740 params->xmit_policy = xmit_hashtype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004741 params->miimon = miimon;
Ben Hutchingsad246c92011-04-26 15:25:52 +00004742 params->num_peer_notif = num_peer_notif;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004743 params->arp_interval = arp_interval;
Jay Vosburghf5b2b962006-09-22 21:54:53 -07004744 params->arp_validate = arp_validate_value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004745 params->updelay = updelay;
4746 params->downdelay = downdelay;
4747 params->use_carrier = use_carrier;
4748 params->lacp_fast = lacp_fast;
4749 params->primary[0] = 0;
Jiri Pirkoa5499522009-09-25 03:28:09 +00004750 params->primary_reselect = primary_reselect_value;
Jay Vosburgh3915c1e82008-05-17 21:10:14 -07004751 params->fail_over_mac = fail_over_mac_value;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +00004752 params->tx_queues = tx_queues;
Andy Gospodarekebd8e492010-06-02 08:39:21 +00004753 params->all_slaves_active = all_slaves_active;
Flavio Leitnerc2952c32010-10-05 14:23:59 +00004754 params->resend_igmp = resend_igmp;
stephen hemminger655f8912011-06-22 09:54:39 +00004755 params->min_links = min_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004756
4757 if (primary) {
4758 strncpy(params->primary, primary, IFNAMSIZ);
4759 params->primary[IFNAMSIZ - 1] = 0;
4760 }
4761
4762 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4763
4764 return 0;
4765}
4766
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004767static struct lock_class_key bonding_netdev_xmit_lock_key;
David S. Millercf508b12008-07-22 14:16:42 -07004768static struct lock_class_key bonding_netdev_addr_lock_key;
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004769
David S. Millere8a04642008-07-17 00:34:19 -07004770static void bond_set_lockdep_class_one(struct net_device *dev,
4771 struct netdev_queue *txq,
4772 void *_unused)
David S. Millerc773e842008-07-08 23:13:53 -07004773{
4774 lockdep_set_class(&txq->_xmit_lock,
4775 &bonding_netdev_xmit_lock_key);
4776}
4777
4778static void bond_set_lockdep_class(struct net_device *dev)
4779{
David S. Millercf508b12008-07-22 14:16:42 -07004780 lockdep_set_class(&dev->addr_list_lock,
4781 &bonding_netdev_addr_lock_key);
David S. Millere8a04642008-07-17 00:34:19 -07004782 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
David S. Millerc773e842008-07-08 23:13:53 -07004783}
4784
Stephen Hemminger181470f2009-06-12 19:02:52 +00004785/*
4786 * Called from registration process
4787 */
4788static int bond_init(struct net_device *bond_dev)
4789{
4790 struct bonding *bond = netdev_priv(bond_dev);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004791 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
Neil Horman9fe06172011-05-25 08:13:01 +00004792 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
Stephen Hemminger181470f2009-06-12 19:02:52 +00004793
4794 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4795
Neil Horman9fe06172011-05-25 08:13:01 +00004796 /*
4797 * Initialize locks that may be required during
4798 * en/deslave operations. All of the bond_open work
4799 * (of which this is part) should really be moved to
4800 * a phase prior to dev_open
4801 */
4802 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4803 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4804
Stephen Hemminger181470f2009-06-12 19:02:52 +00004805 bond->wq = create_singlethread_workqueue(bond_dev->name);
4806 if (!bond->wq)
4807 return -ENOMEM;
4808
4809 bond_set_lockdep_class(bond_dev);
4810
Stephen Hemminger181470f2009-06-12 19:02:52 +00004811 bond_create_proc_entry(bond);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004812 list_add_tail(&bond->bond_list, &bn->dev_list);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004813
Eric W. Biederman6151b3d2009-10-29 14:18:22 +00004814 bond_prepare_sysfs_group(bond);
Jiri Pirko22bedad32010-04-01 21:22:57 +00004815
Taku Izumif073c7c2010-12-09 15:17:13 +00004816 bond_debug_register(bond);
4817
Jiri Pirko22bedad32010-04-01 21:22:57 +00004818 __hw_addr_init(&bond->mc_list);
Stephen Hemminger181470f2009-06-12 19:02:52 +00004819 return 0;
4820}
4821
Eric W. Biederman88ead972009-10-29 14:18:25 +00004822static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4823{
4824 if (tb[IFLA_ADDRESS]) {
4825 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4826 return -EINVAL;
4827 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4828 return -EADDRNOTAVAIL;
4829 }
4830 return 0;
4831}
4832
Jiri Pirkod5da4512011-08-10 06:09:44 +00004833static int bond_get_tx_queues(struct net *net, struct nlattr *tb[],
4834 unsigned int *num_queues,
4835 unsigned int *real_num_queues)
4836{
4837 *num_queues = tx_queues;
4838 return 0;
4839}
4840
Eric W. Biederman88ead972009-10-29 14:18:25 +00004841static struct rtnl_link_ops bond_link_ops __read_mostly = {
4842 .kind = "bond",
Eric W. Biederman66391042009-10-29 23:58:54 +00004843 .priv_size = sizeof(struct bonding),
Eric W. Biederman88ead972009-10-29 14:18:25 +00004844 .setup = bond_setup,
4845 .validate = bond_validate,
Jiri Pirkod5da4512011-08-10 06:09:44 +00004846 .get_tx_queues = bond_get_tx_queues,
Eric W. Biederman88ead972009-10-29 14:18:25 +00004847};
4848
Mitch Williamsdfe60392005-11-09 10:36:04 -08004849/* Create a new bond based on the specified name and bonding parameters.
Jay Vosburghe4b91c42007-01-19 18:15:31 -08004850 * If name is NULL, obtain a suitable "bond%d" name for us.
Mitch Williamsdfe60392005-11-09 10:36:04 -08004851 * Caller must NOT hold rtnl_lock; we need to release it here before we
4852 * set up our sysfs entries.
4853 */
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004854int bond_create(struct net *net, const char *name)
Mitch Williamsdfe60392005-11-09 10:36:04 -08004855{
4856 struct net_device *bond_dev;
4857 int res;
4858
4859 rtnl_lock();
Jay Vosburgh027ea042008-01-17 16:25:02 -08004860
Jiri Pirko1c5cae82011-04-30 01:21:32 +00004861 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4862 name ? name : "bond%d",
4863 bond_setup, tx_queues);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004864 if (!bond_dev) {
Joe Perchesa4aee5c2009-12-13 20:06:07 -08004865 pr_err("%s: eek! can't alloc netdev!\n", name);
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004866 rtnl_unlock();
4867 return -ENOMEM;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004868 }
4869
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004870 dev_net_set(bond_dev, net);
Eric W. Biederman88ead972009-10-29 14:18:25 +00004871 bond_dev->rtnl_link_ops = &bond_link_ops;
4872
Mitch Williamsdfe60392005-11-09 10:36:04 -08004873 res = register_netdevice(bond_dev);
Peter Zijlstra0daa23032006-11-08 19:51:01 -08004874
Phil Oestere826eaf2011-03-14 06:22:05 +00004875 netif_carrier_off(bond_dev);
4876
Mitch Williamsdfe60392005-11-09 10:36:04 -08004877 rtnl_unlock();
Amerigo Wang9e2e61f2010-03-31 21:30:52 +00004878 if (res < 0)
4879 bond_destructor(bond_dev);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004880 return res;
4881}
4882
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004883static int __net_init bond_net_init(struct net *net)
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004884{
Eric W. Biederman15449742009-11-29 15:46:04 +00004885 struct bond_net *bn = net_generic(net, bond_net_id);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004886
4887 bn->net = net;
4888 INIT_LIST_HEAD(&bn->dev_list);
4889
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004890 bond_create_proc_dir(bn);
Eric W. Biederman15449742009-11-29 15:46:04 +00004891
4892 return 0;
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004893}
4894
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004895static void __net_exit bond_net_exit(struct net *net)
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004896{
Eric W. Biederman15449742009-11-29 15:46:04 +00004897 struct bond_net *bn = net_generic(net, bond_net_id);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004898
4899 bond_destroy_proc_dir(bn);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004900}
4901
4902static struct pernet_operations bond_net_ops = {
4903 .init = bond_net_init,
4904 .exit = bond_net_exit,
Eric W. Biederman15449742009-11-29 15:46:04 +00004905 .id = &bond_net_id,
4906 .size = sizeof(struct bond_net),
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004907};
4908
Linus Torvalds1da177e2005-04-16 15:20:36 -07004909static int __init bonding_init(void)
4910{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004911 int i;
4912 int res;
4913
Amerigo Wangbd33acc2011-03-06 21:58:46 +00004914 pr_info("%s", bond_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004915
Mitch Williamsdfe60392005-11-09 10:36:04 -08004916 res = bond_check_params(&bonding_defaults);
Stephen Hemminger3d632c32009-06-12 19:02:48 +00004917 if (res)
Mitch Williamsdfe60392005-11-09 10:36:04 -08004918 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004919
Eric W. Biederman15449742009-11-29 15:46:04 +00004920 res = register_pernet_subsys(&bond_net_ops);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004921 if (res)
4922 goto out;
Jay Vosburgh027ea042008-01-17 16:25:02 -08004923
Eric W. Biederman88ead972009-10-29 14:18:25 +00004924 res = rtnl_link_register(&bond_link_ops);
4925 if (res)
Eric W. Biederman66391042009-10-29 23:58:54 +00004926 goto err_link;
Eric W. Biederman88ead972009-10-29 14:18:25 +00004927
Taku Izumif073c7c2010-12-09 15:17:13 +00004928 bond_create_debugfs();
4929
Linus Torvalds1da177e2005-04-16 15:20:36 -07004930 for (i = 0; i < max_bonds; i++) {
Eric W. Biedermanec87fd32009-10-29 14:18:26 +00004931 res = bond_create(&init_net, NULL);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004932 if (res)
4933 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004934 }
4935
Mitch Williamsb76cdba2005-11-09 10:36:41 -08004936 res = bond_create_sysfs();
4937 if (res)
4938 goto err;
4939
Linus Torvalds1da177e2005-04-16 15:20:36 -07004940 register_netdevice_notifier(&bond_netdev_notifier);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04004941 register_inetaddr_notifier(&bond_inetaddr_notifier);
Mitch Williamsdfe60392005-11-09 10:36:04 -08004942out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004943 return res;
Eric W. Biederman88ead972009-10-29 14:18:25 +00004944err:
4945 rtnl_link_unregister(&bond_link_ops);
Eric W. Biederman66391042009-10-29 23:58:54 +00004946err_link:
Eric W. Biederman15449742009-11-29 15:46:04 +00004947 unregister_pernet_subsys(&bond_net_ops);
Eric W. Biederman88ead972009-10-29 14:18:25 +00004948 goto out;
Mitch Williamsdfe60392005-11-09 10:36:04 -08004949
Linus Torvalds1da177e2005-04-16 15:20:36 -07004950}
4951
4952static void __exit bonding_exit(void)
4953{
4954 unregister_netdevice_notifier(&bond_netdev_notifier);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04004955 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004956
Pavel Emelyanovae68c392008-05-02 17:49:39 -07004957 bond_destroy_sysfs();
Taku Izumif073c7c2010-12-09 15:17:13 +00004958 bond_destroy_debugfs();
Pavel Emelyanovae68c392008-05-02 17:49:39 -07004959
Eric W. Biederman88ead972009-10-29 14:18:25 +00004960 rtnl_link_unregister(&bond_link_ops);
Eric W. Biederman15449742009-11-29 15:46:04 +00004961 unregister_pernet_subsys(&bond_net_ops);
Neil Hormane843fa52010-10-13 16:01:50 +00004962
4963#ifdef CONFIG_NET_POLL_CONTROLLER
Neil Hormanfb4fa762010-12-06 09:05:50 +00004964 /*
4965 * Make sure we don't have an imbalance on our netpoll blocking
4966 */
4967 WARN_ON(atomic_read(&netpoll_block_tx));
Neil Hormane843fa52010-10-13 16:01:50 +00004968#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004969}
4970
4971module_init(bonding_init);
4972module_exit(bonding_exit);
4973MODULE_LICENSE("GPL");
4974MODULE_VERSION(DRV_VERSION);
4975MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4976MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
Eric W. Biederman88ead972009-10-29 14:18:25 +00004977MODULE_ALIAS_RTNL_LINK("bond");